ui: spin the compaction icon while the summary runs

The divider reported the phase in its title while the icon still showed a finished block, so a run in progress and a summary that had landed looked the same in the one place the eye reads first. The icon spins while the summary is running and folds back once it is in.
ui: open a compaction summary at the top of the screen
2026-10-30 16:37:36 -05:00 · 2026-10-21 10:38:23 +02:00 · 2026-10-21 10:27:29 +02:00 · 2026-10-21 10:27:29 +02:00 · 2026-10-21 10:27:29 +02:00 · 2026-10-21 10:27:29 +02:00
690 changed files with 46004 additions and 103453 deletions
@@ -1,172 +0,0 @@
-name: Behavior Check
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/behavior-check.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.py',
-      '**/requirements*.txt',
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/behavior-check.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.py',
-      '**/requirements*.txt',
-    ]
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  GGML_SCHED_DEBUG_REALLOC: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  ubuntu:
-    runs-on: ubuntu-26.04-arm
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          restore: false
-          save: false
-
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          sudo apt update
-          sudo apt install -y build-essential libssl-dev git-lfs libjpeg-dev python3 python3-venv python3-pip
-
-      - name: Python Dependencies
-        id: python_depends
-        run: |
-          python3 -m venv .venv
-          source .venv/bin/activate
-          pip install -r requirements/requirements-convert_hf_to_gguf.txt
-
-      - name: ccache-buckets-restore
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
-        with:
-          key: behavior-check-ubuntu-26.04-arm
-          folder: llama.cpp
-          hf_bucket: ggml-org/cache
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DGGML_NATIVE=OFF \
-            -DGGML_BACKEND_DL=OFF \
-            -DLLAMA_FATAL_WARNINGS=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: ccache-buckets-save
-        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-        uses: ./.github/actions/ccache-buckets
-        env:
-          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
-        with:
-          key: behavior-check-ubuntu-26.04-arm
-          folder: llama.cpp
-          evict-old-files: 1d
-          hf_bucket: ggml-org/cache
-          save: true
-
-      - name: Test
-        id: cmake_test
-        run: |
-          source .venv/bin/activate
-          cd build
-          ctest -L behavior -LE 'model|backend' --verbose --timeout 900
-
-  windows:
-    runs-on: windows-2025
-
-    cache-mode: write
-    permissions:
-      actions: write
-      contents: read
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.24
-        with:
-          key: behavior-check-windows-2025
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Python Dependencies
-        id: python_depends
-        shell: bash
-        run: |
-          python3 -m venv .venv
-          source .venv/Scripts/activate
-          pip install -r requirements/requirements-convert_hf_to_gguf.txt
-
-      - name: Build
-        id: cmake_build
-        shell: cmd
-        run: |
-          call "C:\Program Files\Microsoft Visual Studio\18\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
-          cmake -B build -G "Ninja Multi-Config" ^
-            -DCMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake ^
-            -DCMAKE_BUILD_TYPE=Release ^
-            -DGGML_NATIVE=OFF ^
-            -DGGML_BACKEND_DL=OFF
-          set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
-          cmake --build build --config Release -j %NINJA_JOBS%
-
-      - name: Test
-        id: cmake_test
-        shell: bash
-        run: |
-          source .venv/Scripts/activate
-          cd build
-          ctest -C Release -L behavior -LE 'model|backend' --verbose --timeout 900
-
-      - name: ccache-clear
-        uses: ./.github/actions/ccache-clear
-        env:
-          GH_TOKEN: ${{ github.token }}
-        with:
-          key: behavior-check-windows-2025
-          older: 5m
-          min: 1
-          dry-run: ${{ github.event_name != 'push' || github.ref != 'refs/heads/master' }}
@@ -1,160 +0,0 @@
-name: CI (apple - daily)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  schedule:
-    - cron: '0 0 * * *'
-
-cache-mode: none
-permissions:
-  contents: read
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  macos-latest-ios-xcode:
-    runs-on: macos-latest
-
-    permissions:
-      actions: write
-      contents: read
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-
-      - name: Setup Xcode
-        uses: ggml-org/setup-xcode@v1
-        with:
-          xcode-version: latest-stable
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          cmake -B build -G Xcode \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DLLAMA_BUILD_APP=OFF \
-            -DLLAMA_BUILD_EXAMPLES=OFF \
-            -DLLAMA_BUILD_TOOLS=OFF \
-            -DLLAMA_BUILD_TESTS=OFF \
-            -DLLAMA_BUILD_SERVER=OFF \
-            -DCMAKE_SYSTEM_NAME=iOS \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
-            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
-          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
-
-      - name: xcodebuild for swift package
-        id: xcodebuild
-        run: |
-          ./build-xcframework.sh
-
-      - name: Upload xcframework artifact
-        uses: actions/upload-artifact@v6
-        with:
-          name: llama-xcframework
-          path: build-apple/llama.xcframework/
-          retention-days: 1
-
-      - name: Build Xcode project
-        run: |
-          xcodebuild -downloadPlatform iOS
-          xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' FRAMEWORK_FOLDER_PATH=./build-ios build
-
-  macos-latest-tvos:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          cmake -B build -G Xcode \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DLLAMA_BUILD_COMMON=OFF \
-            -DLLAMA_BUILD_APP=OFF \
-            -DLLAMA_BUILD_EXAMPLES=OFF \
-            -DLLAMA_BUILD_TOOLS=OFF \
-            -DLLAMA_BUILD_TESTS=OFF \
-            -DLLAMA_BUILD_SERVER=OFF \
-            -DCMAKE_SYSTEM_NAME=tvOS \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
-            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
-          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
-
-  macos-latest-visionos:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          cmake -B build -G Xcode \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DLLAMA_BUILD_COMMON=OFF \
-            -DLLAMA_BUILD_APP=OFF \
-            -DLLAMA_BUILD_EXAMPLES=OFF \
-            -DLLAMA_BUILD_TOOLS=OFF \
-            -DLLAMA_BUILD_TESTS=OFF \
-            -DLLAMA_BUILD_SERVER=OFF \
-            -DCMAKE_SYSTEM_NAME=visionOS \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=1.0 \
-            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
-          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
-
-  macos-latest-swift:
-    runs-on: macos-latest
-    needs: macos-latest-ios-xcode
-
-    permissions:
-      actions: read
-      contents: read
-
-    strategy:
-      matrix:
-        destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Download xcframework artifact
-        uses: actions/download-artifact@v7
-        with:
-          name: llama-xcframework
-          path: build-apple/llama.xcframework/
-
-      - name: Build llama.cpp with CMake
-        id: cmake_build
-        run: |
-          sysctl -a
-          cmake -B build -G Xcode \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DLLAMA_BUILD_APP=OFF \
-            -DLLAMA_BUILD_EXAMPLES=OFF \
-            -DLLAMA_BUILD_TOOLS=OFF \
-            -DLLAMA_BUILD_TESTS=OFF \
-            -DLLAMA_BUILD_SERVER=OFF \
-            -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"
-          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
@@ -105,5 +105,198 @@ jobs:
          cd build
          # Metal Paravirtual devices are difficult to support -> disable
          # ref: https://github.com/ggml-org/llama.cpp/pull/19802#issuecomment-4013704023
-          ctest -L main -E "test-llama-archs|test-llama-context|test-speculative-rejection-sampling" --verbose --timeout 900
+          ctest -L main -E "test-llama-archs|test-llama-context" --verbose --timeout 900

+  macos-latest-x64:
+    runs-on: macos-15-intel
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.24
+        with:
+          restore: false
+          save: false
+
+      - name: ccache-buckets-restore
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
+        with:
+          key: apple-x64
+          folder: llama.cpp
+          hf_bucket: ggml-org/cache
+
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          # Metal is disabled due to intermittent failures with Github runners not having a GPU:
+          # https://github.com/ggml-org/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313
+          cmake -B build \
+            -DCMAKE_BUILD_RPATH="@loader_path" \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_BUILD_BORINGSSL=ON \
+            -DGGML_METAL=OFF \
+            -DGGML_RPC=ON \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3
+          time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
+
+      - name: ccache-buckets-save
+        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+        uses: ./.github/actions/ccache-buckets
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}
+        with:
+          key: apple-x64
+          folder: llama.cpp
+          evict-old-files: 1d
+          hf_bucket: ggml-org/cache
+          save: true
+
+      - name: Test
+        id: cmake_test
+        run: |
+          cd build
+          ctest -L main --verbose --timeout 900
+
+  macos-latest-ios-xcode:
+    runs-on: macos-latest
+
+    permissions:
+      actions: write
+      contents: read
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+
+      - name: Setup Xcode
+        uses: ggml-org/setup-xcode@v1
+        with:
+          xcode-version: latest-stable
+
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          cmake -B build -G Xcode \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_OPENSSL=OFF \
+            -DLLAMA_BUILD_APP=OFF \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_TOOLS=OFF \
+            -DLLAMA_BUILD_TESTS=OFF \
+            -DLLAMA_BUILD_SERVER=OFF \
+            -DCMAKE_SYSTEM_NAME=iOS \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
+            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+
+      - name: xcodebuild for swift package
+        id: xcodebuild
+        run: |
+          ./build-xcframework.sh
+
+      - name: Upload xcframework artifact
+        uses: actions/upload-artifact@v6
+        with:
+          name: llama-xcframework
+          path: build-apple/llama.xcframework/
+          retention-days: 1
+
+      - name: Build Xcode project
+        run: |
+          xcodebuild -downloadPlatform iOS
+          xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' FRAMEWORK_FOLDER_PATH=./build-ios build
+
+  macos-latest-tvos:
+    runs-on: macos-latest
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          cmake -B build -G Xcode \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_COMMON=OFF \
+            -DLLAMA_BUILD_APP=OFF \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_TOOLS=OFF \
+            -DLLAMA_BUILD_TESTS=OFF \
+            -DLLAMA_BUILD_SERVER=OFF \
+            -DCMAKE_SYSTEM_NAME=tvOS \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
+            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+
+  macos-latest-visionos:
+    runs-on: macos-latest
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          cmake -B build -G Xcode \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_COMMON=OFF \
+            -DLLAMA_BUILD_APP=OFF \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_TOOLS=OFF \
+            -DLLAMA_BUILD_TESTS=OFF \
+            -DLLAMA_BUILD_SERVER=OFF \
+            -DCMAKE_SYSTEM_NAME=visionOS \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=1.0 \
+            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+
+  macos-latest-swift:
+    runs-on: macos-latest
+    needs: macos-latest-ios-xcode
+
+    permissions:
+      actions: read
+      contents: read
+
+    strategy:
+      matrix:
+        destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: Download xcframework artifact
+        uses: actions/download-artifact@v7
+        with:
+          name: llama-xcframework
+          path: build-apple/llama.xcframework/
+
+      - name: Build llama.cpp with CMake
+        id: cmake_build
+        run: |
+          sysctl -a
+          cmake -B build -G Xcode \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_OPENSSL=OFF \
+            -DLLAMA_BUILD_APP=OFF \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_TOOLS=OFF \
+            -DLLAMA_BUILD_TESTS=OFF \
+            -DLLAMA_BUILD_SERVER=OFF \
+            -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
@@ -9,7 +9,7 @@ permissions:

 jobs:
  linux:
-    runs-on: [self-hosted, Linux, fast]
+    runs-on: [self-hosted, Linux, CPU]
    steps:
      - uses: actions/checkout@v6
        with:
@@ -129,7 +129,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          ctest -L main --verbose --timeout 900
+          ctest -L 'main|python' --verbose --timeout 900

      - name: Test llama2c conversion
        id: llama2c_test
@@ -43,9 +43,9 @@ env:
  LLAMA_ARG_LOG_TIMESTAMPS: 1

 jobs:
-  ubuntu-26-s390x:
-    name: ubuntu-26-s390x (VXE ${{ matrix.vxe }})
-    runs-on: ubuntu-26.04-s390x
+  ubuntu-24-s390x:
+    name: ubuntu-24-s390x (VXE ${{ matrix.vxe }})
+    runs-on: ubuntu-24.04-s390x

    strategy:
      fail-fast: false
@@ -66,6 +66,12 @@ jobs:
            libjpeg-dev build-essential libssl-dev \
            git-lfs

+      - name: Toolchain workaround (GCC 14)
+        run: |
+          sudo apt-get install -y gcc-14 g++-14
+          echo "CC=gcc-14" >> "$GITHUB_ENV"
+          echo "CXX=g++-14" >> "$GITHUB_ENV"
+
      - name: Python Dependencies
        id: python_depends
        run: |
@@ -110,19 +116,30 @@ jobs:

  ubuntu-26-zdnn-s390x:
    name: ubuntu-26-zdnn-s390x
-    runs-on: ubuntu-26.04-s390x
+    runs-on: ubuntu-24.04-s390x
+    container: ubuntu:26.04  # required to get GCC 15.1 and binutils 2.44
+    defaults:
+      run:
+        shell: bash

    steps:
+      - name: Build Dependencies
+        id: build_depends
+        run: |
+          apt-get update
+          apt-get install -y --no-install-recommends \
+            build-essential cmake git ca-certificates \
+            libssl-dev libzdnn-dev
+
      - name: Clone
        id: checkout
        uses: actions/checkout@v6

-      - name: Build Dependencies
-        id: build_depends
+      - name: Toolchain workaround (GCC 15)
        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends \
-            build-essential libssl-dev libzdnn-dev
+          apt-get install -y gcc-15 g++-15
+          echo "CC=gcc-15" >> "$GITHUB_ENV"
+          echo "CXX=g++-15" >> "$GITHUB_ENV"

      - name: Build with zDNN Backend
        id: cmake_build
@@ -49,7 +49,7 @@ jobs:
        - sanitizer: THREAD
          machine: ubuntu-24.04
        - sanitizer: UNDEFINED
-          machine: [self-hosted, X64, Linux, CPU]
+          machine: [self-hosted, X64, Linux]

    runs-on: ${{ matrix.machine }}

@@ -46,7 +46,7 @@ env:

 jobs:
  cpu-x64-high-perf:
-    runs-on: [self-hosted, Linux, X64, CPU]
+    runs-on: [self-hosted, Linux, X64]

    steps:
      - name: Clone
@@ -56,7 +56,7 @@ jobs:
      - name: Test
        id: ggml-ci
        run: |
-          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_HIGH_PERF=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_HIGH_PERF=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp

  cpu-arm64-high-perf-graviton4:
    runs-on: ah-ubuntu_24_04-c8g_8x
@@ -60,4 +60,4 @@ jobs:
      - name: Test
        id: ggml-ci
        run: |
-          GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+          GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -79,7 +79,7 @@ jobs:
        id: ggml-ci
        run: |
          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp

      # - name: ccache-buckets-save
      #   if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
@@ -124,7 +124,7 @@ jobs:
        id: ggml-ci
        run: |
          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp

      # - name: ccache-buckets-save
      #   if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
@@ -150,7 +150,7 @@ jobs:
        id: ggml-ci
        run: |
          vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp

  gpu-vulkan-intel-linux:
    runs-on: [self-hosted, Linux, Intel]
@@ -58,12 +58,7 @@ jobs:
      - name: Install dependencies
        run: |
          sudo apt update
-          sudo apt install -y software-properties-common
-          sudo add-apt-repository ppa:deadsnakes/ppa
-          sudo apt update
-          sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan1 mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 libssl-dev time unzip wget python3.12 python3.12-venv
-          sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1
-          sudo update-alternatives --config python3
+          sudo apt install -y build-essential cmake libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan1 mesa-vulkan-drivers libglvnd0 libgl1 libglx0 libegl1 libgles2 libssl-dev time unzip wget python3 python3-venv python3-pip

      - name: ccache
        uses: ggml-org/ccache-action@v1.2.24
@@ -136,4 +131,4 @@ jobs:
        id: ggml-ci
        run: |
          GG_BUILD_WEBGPU=1 GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
-            bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp
+            bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
@@ -172,8 +172,6 @@ jobs:
            - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.tar.gz)
            - [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-arm64.tar.gz)
            - [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-s390x.tar.gz)
-            - [Ubuntu 26.04 s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-26.04-s390x.tar.gz)
-            - [Ubuntu 26.04 s390x (zDNN)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-zdnn-s390x.tar.gz)
            - [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.tar.gz)
            - [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-arm64.tar.gz)
            - [Ubuntu x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-12.8-x64.tar.gz) - [CUDA 12.8 libraries](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-${{ steps.tag.outputs.name }}-bin-ubuntu-cuda-12.8-x64.tar.gz)
@@ -180,8 +180,6 @@ jobs:
            os: ubuntu-24.04-arm
          - build: 's390x'
            os: ubuntu-24.04-s390x
-          - build: '26.04-s390x'
-            os: ubuntu-26.04-s390x

    runs-on: ${{ matrix.os }}

@@ -202,7 +200,7 @@ jobs:
        id: depends
        run: |
          sudo apt-get update
-          sudo apt-get install -y build-essential libssl-dev
+          sudo apt-get install build-essential libssl-dev

      - name: Toolchain workaround (GCC 14)
        if: ${{ contains(matrix.os, 'ubuntu-24.04') }}
@@ -212,7 +210,7 @@ jobs:
          echo "CXX=g++-14" >> "$GITHUB_ENV"

      - name: ccache
-        if: ${{ !contains(matrix.os, 's390x') }}
+        if: ${{ matrix.build != 's390x' }}
        uses: ggml-org/ccache-action@v1.2.24
        with:
          key: release-${{ matrix.os }}-cpu
@@ -244,7 +242,7 @@ jobs:
          archive: false

      - name: ccache-clear
-        if: ${{ !contains(matrix.os, 's390x') }}
+        if: ${{ matrix.build != 's390x' }}
        uses: ./.github/actions/ccache-clear
        with:
          key: release-${{ matrix.os }}-cpu
@@ -970,11 +968,6 @@ jobs:
          evict-old-files: 1d
          max-size: "1G"

-      - name: Tune ccache for ROCm
-        run: |
-          # since ROCm 10.1, clang -E output has no GPU code, so the preprocessor fallback can reuse objects with old GPU code
-          ccache --set-config=depend_mode=true
-
      # - name: Cache ROCm Installation
      #   id: cache-rocm
      #   uses: actions/cache@v5
@@ -1552,8 +1545,6 @@ jobs:
          # builds hit.
          ccache --set-config=compiler_check=content
          ccache --set-config=sloppiness=time_macros,include_file_mtime,include_file_ctime
-          # since ROCm 10.1, clang -E output has no GPU code, so the preprocessor fallback can reuse objects with old GPU code
-          ccache --set-config=depend_mode=true

      - name: Dependencies
        id: depends
@@ -1628,60 +1619,6 @@ jobs:
        with:
          key: release-ubuntu-24.04-rocm-${{ matrix.ROCM_VERSION }}-${{ matrix.build }}

-  ubuntu-26-zdnn:
-    needs: [check-release, ui-build]
-    if: ${{ needs.check-release.outputs.should_release == 'true' }}
-
-    runs-on: ubuntu-26.04-s390x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-
-      - name: Download UI build
-        uses: actions/download-artifact@v8
-        with:
-          name: llama-ui.zip
-          path: tools/ui/dist
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends \
-            build-essential libssl-dev libzdnn-dev
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_INSTALL_RPATH='$ORIGIN' \
-            -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \
-            -DGGML_BACKEND_DL=ON \
-            -DGGML_NATIVE=OFF \
-            -DGGML_VXE=ON \
-            -DGGML_ZDNN=ON \
-            -DCMAKE_C_FLAGS="-march=arch15" \
-            -DCMAKE_CXX_FLAGS="-march=arch15" \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            ${{ env.CMAKE_ARGS }}
-          cmake --build build --config Release -j $(nproc)
-
-      - name: Pack artifacts
-        id: pack_artifacts
-        run: |
-          cp LICENSE ./build/bin/
-          tar -czvf llama-${{ needs.check-release.outputs.tag_name }}-bin-ubuntu-zdnn-s390x.tar.gz --transform "s,^\.,llama-${{ needs.check-release.outputs.tag_name }}," -C ./build/bin .
-
-      - name: Upload artifacts
-        uses: actions/upload-artifact@v7
-        with:
-          path: llama-${{ needs.check-release.outputs.tag_name }}-bin-ubuntu-zdnn-s390x.tar.gz
-          archive: false
-
  ios-xcode:
    needs: [check-release]
    if: ${{ needs.check-release.outputs.should_release == 'true' }}
@@ -4,7 +4,7 @@ include(CheckIncludeFileCXX)

 ### llama.cpp version
 set(LLAMA_VERSION_MAJOR 0)
-set(LLAMA_VERSION_MINOR 8)
+set(LLAMA_VERSION_MINOR 7)
 set(LLAMA_VERSION_PATCH 0)
 set(LLAMA_VERSION_BASE "${LLAMA_VERSION_MAJOR}.${LLAMA_VERSION_MINOR}.${LLAMA_VERSION_PATCH}")

@@ -103,7 +103,7 @@ endif()

 # subprocess spawning isn't a supported/sandbox-friendly operation on mobile OSes or in WASM
 if (CMAKE_SYSTEM_NAME STREQUAL "iOS" OR CMAKE_SYSTEM_NAME STREQUAL "Android" OR ANDROID
-        OR CMAKE_SYSTEM_NAME STREQUAL "Emscripten" OR EMSCRIPTEN OR CMAKE_SYSTEM_NAME STREQUAL "WASI")
+        OR CMAKE_SYSTEM_NAME STREQUAL "Emscripten" OR EMSCRIPTEN)
    set(LLAMA_SUBPROCESS_DEFAULT OFF)
 else()
    set(LLAMA_SUBPROCESS_DEFAULT ON)
@@ -150,7 +150,6 @@ option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured
 # Required for relocatable CMake package
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake)
-include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/win-app.cmake)

 if (NOT DEFINED LLAMA_BUILD_NUMBER)
    set(LLAMA_BUILD_NUMBER        ${BUILD_NUMBER})
@@ -270,8 +269,6 @@ if (LLAMA_BUILD_MTMD AND NOT (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TOOLS))
    add_subdirectory(tools/mtmd)
 endif()

-llama_apply_app_manifests()
-
 #
 # install
 #
@@ -2,29 +2,20 @@
 # multiple collaborators per item can be specified
 #
 # ggml-org/ci               : CISC, danbev, ggerganov, netrunnereve, ngxson, taronaeo
-# ggml-org/ggml-base        : aendk, fairydreaming, ggerganov
 # ggml-org/ggml-cann        : hipudding
-# ggml-org/ggml-cpu         : angt, chaxu01, ggerganov, max-krasnyansky, taronaeo
-# ggml-org/ggml-cuda        : am17an, gaugarg-nv, IMbackK, JohannesGaessler, ORippler, thelittlefireman, ynankani
-# ggml-org/ggml-et          : marty1885
+# ggml-org/ggml-cuda        : JohannesGaessler, am17an, IMbackK, ORippler
 # ggml-org/ggml-hexagon     : lhez, max-krasnyansky
-# ggml-org/ggml-hip         : IMbackK, thelittlefireman
-# ggml-org/ggml-metal       : forforever73, ggerganov, jhen0409, masterFoad
+# ggml-org/ggml-metal       : ggerganov
 # ggml-org/ggml-opencl      : lhez, max-krasnyansky
-# ggml-org/ggml-openvino    : ravi9, wine99
-# ggml-org/ggml-riscv       : xctan
-# ggml-org/ggml-rpc         : am17an, rgerganov
-# ggml-org/ggml-sycl        : arthw, Titaniumtown
+# ggml-org/ggml-rpc         : rgerganov
+# ggml-org/ggml-sycl        : arthw
 # ggml-org/ggml-vulkan      : 0cc4m, jeffbolznv
 # ggml-org/ggml-webgpu      : reeselevine, yomaytk
-# ggml-org/ggml-zdnn        : AlekseiNikiforovIBM, Andreas-Krebbel, taronaeo
-# ggml-org/llama-base       : am17an, CISC, ggerganov, ngxson, sanmai, ServeurpersoCom
-# ggml-org/llama-common     : aldehir, angt, bartowski1182, danbev, ggerganov, ngxson, nikwen, pwilkin, ruixiang63, sanmai
-# ggml-org/llama-examples   : ggerganov
-# ggml-org/llama-models     : CISC, danbev, ggerganov, ngxson
+# ggml-org/ggml-zdnn        : taronaeo
+# ggml-org/llama-common     : ggerganov, aldehir, angt, danbev, ngxson, pwilkin
 # ggml-org/llama-mtmd       : ngxson
-# ggml-org/llama-server     : allozaur, angt, ggerganov, ngxson, ServeurpersoCom
-# ggml-org/llama-ui         : allozaur
+# ggml-org/llama-server     : ggerganov, ngxson, allozaur, angt, ServeurpersoCom
+# ggml-org/llama-ui           : allozaur

 /.devops/*.Dockerfile                   @ngxson
 /.github/actions/                       @ggml-org/ci
@@ -38,12 +29,28 @@
 /conversion/                            @CISC
 /convert_*.py                           @CISC
 /docs/backend/snapdragon/               @ggml-org/ggml-hexagon
-/examples/                              @ggml-org/llama-examples
+/examples/batched.swift/                @ggerganov
+/examples/batched/                      @ggerganov
+/examples/convert-llama2c-to-ggml/      @ggerganov
 /examples/debug/                        @danbev @pwilkin
+/examples/deprecation-warning/          @ggerganov
 /examples/diffusion/                    @am17an
+/examples/embedding/                    @ggerganov
+/examples/eval-callback/                @ggerganov
+/examples/export-docs/                  @ggerganov
+/examples/gen-docs/                     @ggerganov
+/examples/gguf/                         @ggerganov
 /examples/llama.android/                @ggerganov @hanyin-arm @naco-siren
+/examples/llama.swiftui/                @ggerganov
+/examples/llama.vim                     @ggerganov
+/examples/lookahead/                    @ggerganov
 /examples/lookup/                       @JohannesGaessler
 /examples/model-conversion/             @danbev
+/examples/parallel/                     @ggerganov
+/examples/passkey/                      @ggerganov
+/examples/retrieval/                    @ggerganov
+/examples/speculative-simple/           @ggerganov
+/examples/speculative/                  @ggerganov
 /ggml/cmake/                            @ggerganov
 /ggml/include/                          @ggerganov
 /ggml/src/ggml-backend-meta.cpp         @JohannesGaessler
@@ -55,7 +62,7 @@
 /ggml/src/ggml-cuda/                    @ggml-org/ggml-cuda
 /ggml/src/ggml-cuda/vendors/hip.h       @IMbackK
 /ggml/src/ggml-hexagon/                 @ggml-org/ggml-hexagon
-/ggml/src/ggml-hip/                     @ggml-org/ggml-hip
+/ggml/src/ggml-hip/                     @IMbackK
 /ggml/src/ggml-et/                      @marty1885
 /ggml/src/ggml-impl.h                   @ggerganov
 /ggml/src/ggml-metal/                   @ggml-org/ggml-metal
@@ -69,7 +76,7 @@
 /ggml/src/ggml-virtgpu/                 @kpouget
 /ggml/src/ggml-vulkan/                  @ggml-org/ggml-vulkan
 /ggml/src/ggml-webgpu/                  @ggml-org/ggml-webgpu
-/ggml/src/ggml-zdnn/                    @ggml-org/ggml-zdnn
+/ggml/src/ggml-zdnn/                    @ggml-org/ggml-zdnn @Andreas-Krebbel @AlekseiNikiforovIBM
 /ggml/src/ggml-zendnn/                  @avinashcpandey @Jiten1parmar
 /ggml/src/ggml.c                        @ggerganov
 /ggml/src/ggml.cpp                      @ggerganov
@@ -80,14 +87,14 @@
 /scripts/get*                           @ggerganov
 /scripts/sync*                          @ggerganov
 /scripts/snapdragon/                    @ggml-org/ggml-hexagon
-/src/                                   @ggml-org/llama-base
+/src/                                   @ggerganov
 /src/llama-adapter.*                    @CISC
 /src/llama-arch.*                       @CISC
 /src/llama-chat.*                       @ngxson
 /src/llama-graph.*                      @CISC
 /src/llama-model.*                      @CISC
 /src/llama-vocab.*                      @CISC
-/src/models/                            @ggml-org/llama-models
+/src/models/                            @CISC
 /tests/                                 @ggerganov
 /tests/test-chat.*                      @pwilkin
 /tests/test-server-component.cpp        @ggml-org/llama-server
@@ -13,7 +13,7 @@
 [![Docker](https://img.shields.io/github/actions/workflow/status/ggml-org/llama.cpp/docker.yml?label=Docker)](https://github.com/ggml-org/llama.cpp/actions/workflows/docker.yml)
 [![Winget](https://img.shields.io/github/actions/workflow/status/ggml-org/llama.cpp/winget.yml?label=Winget)](https://github.com/ggml-org/llama.cpp/actions/workflows/winget.yml)

-[ggml](https://github.com/ggml-org/ggml) / [ops](https://github.com/ggml-org/llama.cpp/blob/master/docs/ops.md) / [maintainer PRs](https://github.com/ggml-org/llama.cpp/issues?q=is%3Apr%20is%3Aopen%20draft%3AFalse%20(author%3Asanmai%20OR%20author%3Argerganov%20OR%20author%3AKitaitiMakoto%20OR%20author%3Adanbev%20OR%20author%3Aaldehir%20OR%20author%3Amax-krasnyansky%20OR%20author%3ACISC%20OR%20author%3Aggerganov%20OR%20author%3Aam17an%20OR%20author%3Ajhen0409%20OR%20author%3Abartowski1182%20OR%20author%3Anikwen%20OR%20author%3Ahipudding%20OR%20author%3Aravi9%20OR%20author%3AServeurpersoCom%20OR%20author%3Athelittlefireman%20OR%20author%3Apwilkin%20OR%20author%3Areeselevine%20OR%20author%3Angxson%20OR%20author%3Ajeffbolznv%20OR%20author%3Amarty1885%20OR%20author%3A0cc4m%20OR%20author%3ATitaniumtown%20OR%20author%3Aangt%20OR%20author%3AIMbackK%20OR%20author%3Aarthw%20OR%20author%3AJohannesGaessler%20OR%20author%3AORippler%20OR%20author%3Aruixiang63%20OR%20author%3Axctan%20OR%20author%3AmasterFoad%20OR%20author%3Aallozaur%20OR%20author%3AAndreas-Krebbel%20OR%20author%3Ayomaytk%20OR%20author%3Aaendk%20OR%20author%3Awine99%20OR%20author%3Agaugarg-nv%20OR%20author%3Ataronaeo%20OR%20author%3Aforforever73%20OR%20author%3Achaxu01%20OR%20author%3Alhez%20OR%20author%3AAlekseiNikiforovIBM%20OR%20author%3Anetrunnereve%20OR%20author%3Afairydreaming%20OR%20author%3Aynankani)%20sort%3Aupdated-desc) / [dev stats](https://github.com/ggml-org/llama.cpp-dev) / [lib llama API](https://github.com/ggml-org/llama.cpp/issues/9289) / [llama-server REST API](https://github.com/ggml-org/llama.cpp/issues/9291)
+[ggml](https://github.com/ggml-org/ggml) / [ops](https://github.com/ggml-org/llama.cpp/blob/master/docs/ops.md) / [maintainer PRs](https://github.com/ggml-org/llama.cpp/issues?q=is%3Apr%20is%3Aopen%20draft%3AFalse%20(author%3Argerganov%20OR%20author%3AKitaitiMakoto%20OR%20author%3Adanbev%20OR%20author%3Aaldehir%20OR%20author%3Amax-krasnyansky%20OR%20author%3ACISC%20OR%20author%3Aggerganov%20OR%20author%3Aam17an%20OR%20author%3Ajhen0409%20OR%20author%3Abartowski1182%20OR%20author%3Anikwen%20OR%20author%3Ahipudding%20OR%20author%3Aravi9%20OR%20author%3AServeurpersoCom%20OR%20author%3Apwilkin%20OR%20author%3Areeselevine%20OR%20author%3Angxson%20OR%20author%3Ajeffbolznv%20OR%20author%3Amarty1885%20OR%20author%3A0cc4m%20OR%20author%3ATitaniumtown%20OR%20author%3Aangt%20OR%20author%3AIMbackK%20OR%20author%3Aarthw%20OR%20author%3AJohannesGaessler%20OR%20author%3AORippler%20OR%20author%3Aruixiang63%20OR%20author%3Axctan%20OR%20author%3Aallozaur%20OR%20author%3Ayomaytk%20OR%20author%3Aaendk%20OR%20author%3Awine99%20OR%20author%3Agaugarg-nv%20OR%20author%3Ataronaeo%20OR%20author%3Aforforever73%20OR%20author%3Alhez%20OR%20author%3Anetrunnereve%20OR%20author%3Afairydreaming)%20sort%3Aupdated-desc) / [dev stats](https://github.com/ggml-org/llama.cpp-dev) / [lib llama API](https://github.com/ggml-org/llama.cpp/issues/9289) / [llama-server REST API](https://github.com/ggml-org/llama.cpp/issues/9291)

 </div>

@@ -185,28 +185,18 @@ fi

 ## helpers

-# download a file unless already cached in $MNT
+# download a file if it does not exist or if it is outdated
 function gg_wget {
    local out=$1
    local url=$2

    local cwd=`pwd`
-    local name=`basename $url`
-    local tmp=.gg-wget.$$.tmp

    mkdir -p $out
    cd $out

-    # cache first: HF artifacts are static, HF sends no Last-Modified for -N
-    if [ ! -s $name ]; then
-        # publish with mv: concurrent runners never read partial files
-        rm -rf $tmp && mkdir $tmp
-        (cd $tmp && wget -nv $url)
-        if [ -s $tmp/$name ]; then
-            mv -f $tmp/$name $name
-        fi
-        rm -rf $tmp
-    fi
+    # should not re-download if file is the same
+    wget -nv -c -N $url

    cd $cwd
 }
@@ -264,7 +254,7 @@ function gg_run_ctest_release {
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

    if [ -z ${GG_BUILD_LOW_PERF} ]; then
-        (time ctest -C Release --output-on-failure -L main ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
+        (time ctest -C Release --output-on-failure -L 'main|python' ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
    else
        (time ctest -C Release --output-on-failure -L main -E test-opt ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
    fi
@@ -322,8 +312,8 @@ function gg_run_test_scripts {

    set -e

-    (cd ./tools/gguf-split && time bash tests.sh "$SRC/build-ci-release/bin" "$OUT/scratch") 2>&1 | tee -a $OUT/${ci}-scripts.log
-    (cd ./tools/quantize   && time bash tests.sh "$SRC/build-ci-release/bin" "$OUT/scratch") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./tools/gguf-split && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./tools/quantize   && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log

    set +e
 }
@@ -384,14 +374,10 @@ function gg_run_qwen3_0_6b {


    gg_wget models-mnt/wikitext/ https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
-
-    rm -rf $OUT/scratch/qwen3-0.6B $OUT/scratch/wikitext
-    mkdir -p $OUT/scratch/qwen3-0.6B
-    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d $OUT/scratch/wikitext/
+    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d models-mnt/wikitext/

    path_models="../models-mnt/qwen3/0.6B"
-    path_local="$OUT/scratch/qwen3-0.6B"
-    path_wiki="$OUT/scratch/wikitext/wikitext-2-raw"
+    path_wiki="../models-mnt/wikitext/wikitext-2-raw"

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -400,15 +386,72 @@ function gg_run_qwen3_0_6b {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf  --outtype f16
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-bf16.gguf --outtype bf16
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf  --outtype f16
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-bf16.gguf --outtype bf16

-    model_f16="${path_local}/ggml-model-f16.gguf"
-    model_bf16="${path_local}/ggml-model-bf16.gguf"
-    model_q4_0="${path_local}/ggml-model-q4_0.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_bf16="${path_models}/ggml-model-bf16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"
+    model_q4_0="${path_models}/ggml-model-q4_0.gguf"
+    model_q4_1="${path_models}/ggml-model-q4_1.gguf"
+    model_q5_0="${path_models}/ggml-model-q5_0.gguf"
+    model_q5_1="${path_models}/ggml-model-q5_1.gguf"
+    model_q2_k="${path_models}/ggml-model-q2_k.gguf"
+    model_q3_k="${path_models}/ggml-model-q3_k.gguf"
+    model_q4_k="${path_models}/ggml-model-q4_k.gguf"
+    model_q5_k="${path_models}/ggml-model-q5_k.gguf"
+    model_q6_k="${path_models}/ggml-model-q6_k.gguf"

    wiki_test="${path_wiki}/wiki.test.raw"

+    ./bin/llama-quantize ${model_bf16} ${model_q8_0} q8_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_0} q4_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_1} q4_1 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_0} q5_0 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_1} q5_1 $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q2_k} q2_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q3_k} q3_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q4_k} q4_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q5_k} q5_k $(nproc)
+    ./bin/llama-quantize ${model_bf16} ${model_q6_k} q6_k $(nproc)
+
+    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
+
+    (time ./bin/llama-completion -no-cnv --model ${model_f16}  -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-completion -no-cnv --model ${model_bf16} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q8_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q2_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q3_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q4_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q5_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-completion -no-cnv --model ${model_q6_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    if [ -z ${GG_BUILD_NO_BF16} ]; then
+        (time ./bin/llama-perplexity --model ${model_bf16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
+    fi
+    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
+
+    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
+    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
+    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
+    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
+
    function check_ppl {
        qnt="$1"
        ppl=$(echo "$2" | grep -oE "[0-9]+\.[0-9]+" | tail -n 1)
@@ -422,56 +465,20 @@ function gg_run_qwen3_0_6b {
        return 0
    }

-    for qnt in q8_0 q4_0 q4_1 q5_0 q5_1 q2_k q3_k q4_k q5_k q6_k; do
-        model="${path_local}/ggml-model-${qnt}.gguf"
-
-        ./bin/llama-quantize ${model_bf16} ${model} ${qnt} $(nproc)
-
-        (time ./bin/llama-completion -no-cnv --model ${model} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-${qnt}.log
-        (time ./bin/llama-perplexity --model ${model} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-${qnt}.log
-
-        # note: ppl > 20.0 for q2_k with this model
-        if [ "${qnt}" != "q2_k" ]; then
-            check_ppl "${qnt}" "$(cat $OUT/${ci}-tg-${qnt}.log | grep "^\[1\]")"
-        fi
-
-        if [ "${qnt}" = "q4_0" ]; then
-            # q4_0 is kept for other tests
-            q4_0="${path_models}/ggml-model-q4_0.gguf"
-            ln ${model} "$q4_0" 2>/dev/null || { cp ${model} $q4_0.tmp.$$ && mv -f $q4_0.tmp.$$ $q4_0; }
-        else
-            rm -f ${model}
-        fi
-    done
-
-    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
-
-    (time ./bin/llama-completion -no-cnv --model ${model_f16}  -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    (time ./bin/llama-completion -no-cnv --model ${model_bf16} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
-
-    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        (time ./bin/llama-perplexity --model ${model_bf16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
-    fi
-
-    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
-
-    # TODO: these tests are extremely slow on WebGPU - re-enable when fixed
-    #       https://github.com/ggml-org/llama.cpp/actions/runs/39639985732
-    if [ -z ${GG_BUILD_WEBGPU} ]; then
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-        (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-    fi
-    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-    (time ./bin/test-llama-context --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-test-llama-context.log
-
    check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")"
    if [ -z ${GG_BUILD_NO_BF16} ]; then
        check_ppl "bf16" "$(cat $OUT/${ci}-tg-bf16.log | grep "^\[1\]")"
    fi
-
-    # scratch is rebuilt on every run; q4_0 is already published to $MNT
-    rm -rf $OUT/scratch/qwen3-0.6B $OUT/scratch/wikitext
+    check_ppl "q8_0" "$(cat $OUT/${ci}-tg-q8_0.log | grep "^\[1\]")"
+    check_ppl "q4_0" "$(cat $OUT/${ci}-tg-q4_0.log | grep "^\[1\]")"
+    check_ppl "q4_1" "$(cat $OUT/${ci}-tg-q4_1.log | grep "^\[1\]")"
+    check_ppl "q5_0" "$(cat $OUT/${ci}-tg-q5_0.log | grep "^\[1\]")"
+    check_ppl "q5_1" "$(cat $OUT/${ci}-tg-q5_1.log | grep "^\[1\]")"
+   #check_ppl "q2_k" "$(cat $OUT/${ci}-tg-q2_k.log | grep "^\[1\]")" # note: ppl > 20.0 for this quant and model
+    check_ppl "q3_k" "$(cat $OUT/${ci}-tg-q3_k.log | grep "^\[1\]")"
+    check_ppl "q4_k" "$(cat $OUT/${ci}-tg-q4_k.log | grep "^\[1\]")"
+    check_ppl "q5_k" "$(cat $OUT/${ci}-tg-q5_k.log | grep "^\[1\]")"
+    check_ppl "q6_k" "$(cat $OUT/${ci}-tg-q6_k.log | grep "^\[1\]")"

    set +e
 }
@@ -494,8 +501,6 @@ function gg_run_embd_bge_small {
    gg_wget models-mnt/bge-small/1_Pooling https://huggingface.co/BAAI/bge-small-en-v1.5/raw/main/1_Pooling/config.json

    path_models="../models-mnt/bge-small"
-    path_local="$OUT/scratch/bge-small"
-    rm -rf $path_local && mkdir -p $path_local

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -504,10 +509,10 @@ function gg_run_embd_bge_small {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf

-    model_f16="${path_local}/ggml-model-f16.gguf"
-    model_q8_0="${path_local}/ggml-model-q8_0.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"

    ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0

@@ -516,8 +521,6 @@ function gg_run_embd_bge_small {
    (time ./bin/llama-embedding --model ${model_f16}  -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
    (time ./bin/llama-embedding --model ${model_q8_0} -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log

-    rm -rf $path_local
-
    set +e
 }

@@ -534,8 +537,6 @@ function gg_run_rerank_tiny {
    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/vocab.json

    path_models="../models-mnt/rerank-tiny"
-    path_local="$OUT/scratch/rerank-tiny"
-    rm -rf $path_local && mkdir -p $path_local

    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release

@@ -544,9 +545,9 @@ function gg_run_rerank_tiny {
    (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time cmake --build . --config Release -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log

-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_local}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf

-    model_f16="${path_local}/ggml-model-f16.gguf"
+    model_f16="${path_models}/ggml-model-f16.gguf"

    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)

@@ -576,8 +577,6 @@ function gg_run_rerank_tiny {
    check_score "rerank score 1" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 1")" "0.00" "0.05" | tee -a $OUT/${ci}-rk-f16.log
    check_score "rerank score 2" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 2")" "0.10" "0.30" | tee -a $OUT/${ci}-rk-f16.log

-    rm -rf $path_local
-
    set +e
 }

@@ -684,11 +683,11 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
    ln -sfn ${mnt_models} ${SRC}/models-mnt

    # Create a fresh python3 venv and enter it
-    if ! python3 -m venv "$OUT/venv"; then
-        echo "Error: Failed to create Python virtual environment at $OUT/venv."
+    if ! python3 -m venv "$MNT/venv"; then
+        echo "Error: Failed to create Python virtual environment at $MNT/venv."
        exit 1
    fi
-    source "$OUT/venv/bin/activate"
+    source "$MNT/venv/bin/activate"

    pip install -r ${SRC}/requirements.txt --disable-pip-version-check
    pip install --editable gguf-py --disable-pip-version-check
@@ -1,66 +0,0 @@
-set(LLAMA_WIN_APP_MANIFEST "${CMAKE_CURRENT_LIST_DIR}/../common/win-app.manifest")
-
-if(WIN32)
-    if(MSVC OR CMAKE_CXX_SIMULATE_ID STREQUAL "MSVC")
-        set(LLAMA_WIN_APP_MSVC_ABI TRUE)
-    else()
-        set(LLAMA_WIN_APP_MSVC_ABI FALSE)
-    endif()
-
-    if(NOT LLAMA_WIN_APP_MSVC_ABI)
-        if(CMAKE_VERSION VERSION_LESS 3.21)
-            message(FATAL_ERROR "CMake 3.21 or newer is required to build for Windows with the GNU ABI")
-        endif()
-        enable_language(RC)
-        set(LLAMA_WIN_APP_MANIFEST_RC "${CMAKE_CURRENT_BINARY_DIR}/win-app-manifest.rc")
-        configure_file("${CMAKE_CURRENT_LIST_DIR}/win-app.rc.in" "${LLAMA_WIN_APP_MANIFEST_RC}" @ONLY)
-        set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${LLAMA_WIN_APP_MANIFEST}")
-    endif()
-endif()
-
-function(llama_add_app_manifest target)
-    if(NOT WIN32)
-        return()
-    endif()
-    get_target_property(_aliased ${target} ALIASED_TARGET)
-    if(_aliased)
-        return()
-    endif()
-    get_target_property(_type ${target} TYPE)
-    if(NOT _type STREQUAL "EXECUTABLE")
-        return()
-    endif()
-    if(LLAMA_WIN_APP_MSVC_ABI)
-        target_sources(${target} PRIVATE "${LLAMA_WIN_APP_MANIFEST}")
-    else()
-        if(NOT TARGET llama-win-app-manifest-res)
-            add_library(llama-win-app-manifest-res OBJECT ${LLAMA_WIN_APP_MANIFEST_RC})
-            set_source_files_properties(${LLAMA_WIN_APP_MANIFEST_RC} PROPERTIES OBJECT_DEPENDS "${LLAMA_WIN_APP_MANIFEST}")
-        endif()
-        target_link_libraries(${target} PRIVATE "$<TARGET_OBJECTS:llama-win-app-manifest-res>")
-    endif()
-endfunction()
-
-function(llama_apply_app_manifests_dir dir)
-    get_property(_targets DIRECTORY "${dir}" PROPERTY BUILDSYSTEM_TARGETS)
-    foreach(_target IN LISTS _targets)
-        llama_add_app_manifest(${_target})
-    endforeach()
-    get_property(_subdirs DIRECTORY "${dir}" PROPERTY SUBDIRECTORIES)
-    foreach(_subdir IN LISTS _subdirs)
-        llama_apply_app_manifests_dir("${_subdir}")
-    endforeach()
-endfunction()
-
-function(llama_apply_app_manifests)
-    if(NOT WIN32)
-        return()
-    endif()
-    get_property(_subdirs DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}" PROPERTY SUBDIRECTORIES)
-    foreach(_dir IN LISTS _subdirs)
-        get_filename_component(_name "${_dir}" NAME)
-        if(_name STREQUAL "app" OR _name STREQUAL "examples" OR _name STREQUAL "pocs" OR _name STREQUAL "tests" OR _name STREQUAL "tools")
-            llama_apply_app_manifests_dir("${_dir}")
-        endif()
-    endforeach()
-endfunction()
@@ -1,3 +0,0 @@
-// generated from cmake/win-app.rc.in - do not edit
-// RT_MANIFEST (24), ID 1: the application manifest shared by all llama.cpp executables
-1 24 "@LLAMA_WIN_APP_MANIFEST@"
@@ -78,8 +78,6 @@ add_library(${TARGET}
    download.h
    fit.cpp
    fit.h
-    file-lock.cpp
-    file-lock.h
    hf-cache.cpp
    hf-cache.h
    http.h
@@ -186,8 +184,3 @@ if (LLAMA_LLGUIDANCE)
 endif()

 target_link_libraries(${TARGET} PUBLIC llama Threads::Threads)
-
-if (CMAKE_SYSTEM_NAME MATCHES "AIX")
-    # AIX's flock() function comes from libbsd.a
-    target_link_libraries(${TARGET} PRIVATE -lbsd)
-endif()
@@ -12,6 +12,16 @@
 #include "speculative.h"
 #include "preset.h"

+// fix problem with std::min and std::max
+#if defined(_WIN32)
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
+#include <windows.h>
+#include <shellapi.h>
+#endif
+
 #include <algorithm>
 #include <cinttypes>
 #include <climits>
@@ -28,6 +38,20 @@
 #include <thread> // for hardware_concurrency
 #include <vector>

+#ifndef __EMSCRIPTEN__
+#ifdef __linux__
+#include <linux/limits.h>
+#elif defined(_WIN32)
+#   if !defined(PATH_MAX)
+#   define PATH_MAX MAX_PATH
+#   endif
+#elif defined(_AIX)
+#include <sys/limits.h>
+#else
+#include <sys/syslimits.h>
+#endif
+#endif
+
 #define LLAMA_MAX_URL_LENGTH 2084 // Maximum URL Length in Chrome: 2083

 using json = common_json;
@@ -327,7 +351,7 @@ static bool parse_bool_value(const std::string & value) {
 static std::string get_default_local_path(const std::string & url) {
    auto f = string_split<std::string>(url, '#').front();
    f = string_split<std::string>(f, '?').front();
-    return fs_get_cache_file(string_split<std::string>(f, '/').back()).string();
+    return fs_path_to_utf8(fs_get_cache_file(string_split<std::string>(f, '/').back()));
 }

 static bool spec_types_is_default(const common_params & params) {
@@ -446,14 +470,6 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    auto opts = handler.opts; // copy
    opts.callback = callback;

-    // another process may hold the cache lock during finalization:
-    // let the download callback cancel that wait too
-    auto finalize = [callback](const hf_cache::hf_file & f) {
-        return hf_cache::finalize_file(f, [callback]() {
-            return !(callback && callback->is_cancelled());
-        });
-    };
-
    // handle plain "url" if needed
    auto handle_url = [&](common_params_model & model) {
        if (!model.url.empty()) {
@@ -502,7 +518,7 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    }

    // handle hf_plan tasks
-    auto add_tasks = [&opts, &tasks, &finalize](const hf_cache::hf_files  & model_files,
+    auto add_tasks = [&opts, &tasks](const hf_cache::hf_files  & model_files,
                                    const hf_cache::hf_file    & primary,
                                    common_params_model        & model) {
        for (size_t i = 0; i < model_files.size(); ++i) {
@@ -511,9 +527,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
            tasks.emplace_back(model_file, opts, [&, is_primary]() {
                if (is_primary) {
                    // the primary file is the first split (00001-of), use it as model path
-                    model.path = finalize(model_file);
+                    model.path = hf_cache::finalize_file(model_file);
                } else {
-                    finalize(model_file);
+                    hf_cache::finalize_file(model_file);
                }
            });
        }
@@ -565,9 +581,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.mtp, opts, [&]() {
            // only use the discovered MTP head when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.mtp);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.mtp);
            } else {
-                finalize(plan_spec.mtp);
+                hf_cache::finalize_file(plan_spec.mtp);
            }
        });
    }
@@ -575,9 +591,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.dflash, opts, [&]() {
            // only use the discovered DFlash sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.dflash);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.dflash);
            } else {
-                finalize(plan_spec.dflash);
+                hf_cache::finalize_file(plan_spec.dflash);
            }
        });
    }
@@ -585,9 +601,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.eagle3, opts, [&]() {
            // only use the discovered Eagle3 sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.eagle3);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.eagle3);
            } else {
-                finalize(plan_spec.eagle3);
+                hf_cache::finalize_file(plan_spec.eagle3);
            }
        });
    }
@@ -595,9 +611,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan_spec.dspark, opts, [&]() {
            // only use the discovered DSpark sidecar when no draft path is set yet
            if (params.speculative.draft.mparams.path.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan_spec.dspark);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan_spec.dspark);
            } else {
-                finalize(plan_spec.dspark);
+                hf_cache::finalize_file(plan_spec.dspark);
            }
        });
    }
@@ -618,16 +634,16 @@ void common_models_handler_apply(common_models_handler & handler, common_params
    }
    if (!plan.mmproj.local_path.empty()) {
        tasks.emplace_back(plan.mmproj, opts, [&]() {
-            params.mmproj.path = finalize(plan.mmproj);
+            params.mmproj.path = hf_cache::finalize_file(plan.mmproj);
        });
    }
    if (!plan.mtp.local_path.empty() && !had_spec_url) {
        tasks.emplace_back(plan.mtp, opts, [&]() {
            // only fall back to the discovered MTP head when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.mtp);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.mtp);
            } else {
-                finalize(plan.mtp);
+                hf_cache::finalize_file(plan.mtp);
            }
        });
    }
@@ -635,9 +651,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.dflash, opts, [&]() {
            // only fall back to the discovered DFlash sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.dflash);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.dflash);
            } else {
-                finalize(plan.dflash);
+                hf_cache::finalize_file(plan.dflash);
            }
        });
    }
@@ -645,9 +661,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.eagle3, opts, [&]() {
            // only fall back to the discovered Eagle3 sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.eagle3);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.eagle3);
            } else {
-                finalize(plan.eagle3);
+                hf_cache::finalize_file(plan.eagle3);
            }
        });
    }
@@ -655,9 +671,9 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.dspark, opts, [&]() {
            // only fall back to the discovered DSpark sidecar when no draft was explicitly provided
            if (params.speculative.draft.mparams.empty()) {
-                params.speculative.draft.mparams.path = finalize(plan.dspark);
+                params.speculative.draft.mparams.path = hf_cache::finalize_file(plan.dspark);
            } else {
-                finalize(plan.dspark);
+                hf_cache::finalize_file(plan.dspark);
            }
        });
    }
@@ -665,7 +681,7 @@ void common_models_handler_apply(common_models_handler & handler, common_params
        tasks.emplace_back(plan.preset, opts, [&]() {
            // if HF repo is a preset repo, we simply run server in router mode with the preset.ini file
            params.models_preset_hf = params.model.hf_repo; // only for showing a warning
-            params.models_preset    = finalize(plan.preset);
+            params.models_preset    = hf_cache::finalize_file(plan.preset);
            // clear the model so the server starts in router mode
            params.model.path.clear();
            params.model.hf_repo.clear();
@@ -738,7 +754,7 @@ static void common_params_apply_system_config(common_params & params, llama_exam
    common_preset_context ctx(ex);
    ctx.ignore_unknown_keys = true; // the same config file is shared by all programs
    for (const auto & path : found) {
-        LOG_INF("using config file: %s\n", path.string().c_str());
+        LOG_INF("using config file: %s\n", fs_path_to_utf8(path).c_str());
        common_preset global;
        common_presets presets = ctx.load_from_ini(path, global);
        global.apply_to_params(params);
@@ -900,7 +916,7 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
        // TODO @ngxson : maybe show a list of available models in CLI in this case
        bool can_skip_model = params.usage || params.completion || !params.server_base.empty();
        if (!can_skip_model && params.model.path.empty()) {
-            throw std::invalid_argument("error: --model or -hf is required\n");
+            throw std::invalid_argument("error: --model is required\n");
        }
    }

@@ -1235,7 +1251,44 @@ bool common_params_to_map(int argc, char ** argv, llama_example ex, std::map<com
    return true;
 }

+#ifdef _WIN32
+struct utf8_argv {
+    std::vector<std::string> buf;
+    std::vector<char*> ptrs;
+};
+
+static utf8_argv make_utf8_argv() {
+    utf8_argv out;
+    int wargc = 0;
+    LPWSTR* wargv = CommandLineToArgvW(GetCommandLineW(), &wargc);
+    if (!wargv) return out;
+
+    out.buf.reserve(wargc);
+    for (int i = 0; i < wargc; ++i) {
+        int n = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS, wargv[i], -1, nullptr, 0, nullptr, nullptr);
+        if (n <= 0) { out.buf.emplace_back(); continue; }
+        auto& s = out.buf.emplace_back();
+        s.resize(static_cast<size_t>(n - 1));
+        (void)WideCharToMultiByte(CP_UTF8, 0, wargv[i], -1, s.data(), n, nullptr, nullptr);
+    }
+    LocalFree(wargv);
+
+    out.ptrs.reserve(out.buf.size() + 1);
+    for (auto& s : out.buf) out.ptrs.push_back(s.data());
+    out.ptrs.push_back(nullptr);
+    return out;
+}
+#endif
+
 bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
+#ifdef _WIN32
+    auto utf8 = make_utf8_argv();
+    // repair argv only when it matches the process command line
+    if (static_cast<int>(utf8.buf.size()) == argc) {
+        argv = utf8.ptrs.data();
+    }
+#endif
+
    auto ctx_arg = common_params_parser_init(params, ex, print_usage);
    const common_params params_org = ctx_arg.params; // the example can modify the default params

@@ -3647,13 +3700,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.models_max = value;
        }
    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_MODELS_MAX"));
-    add_opt(common_arg(
-        {"--models-memory-margin"}, "N",
-        string_format("for router server, MiB of device memory to keep free per device, unload LRU models until a new one fits (default: %d, 0 = disabled)", params.models_memory_margin),
-        [](common_params & params, int value) {
-            params.models_memory_margin = value;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_MODELS_MEMORY_MARGIN"));
    add_opt(common_arg(
        {"--models-autoload"},
        {"--no-models-autoload"},
@@ -4451,14 +4497,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
            params.tts_lang = value;
        }
    ).set_examples({LLAMA_EXAMPLE_TTS}));
-    add_opt(common_arg(
-        {"--tts-voice"}, "VOICE",
-        "voice for audio generation (model-dependent)\n"
-        "see tts/README.md for per-model usage notes",
-        [](common_params & params, const std::string & value) {
-            params.tts_voice = value;
-        }
-    ).set_examples({LLAMA_EXAMPLE_TTS}));
    add_opt(common_arg(
        {"--tts-speaker-file"}, "FNAME",
        "speaker file path for audio generation",
@@ -190,17 +190,10 @@ json common_chat_msg::to_json_oaicompat(bool concat_typed_text) const {
        } else {
            auto & parts = jmsg["content"] = json::array();
            for (const auto & part : content_parts) {
-                if (part.type == "tool_reference") {
-                    parts.push_back({
-                        {"type", part.type},
-                        {"name", part.name},
-                    });
-                } else {
-                    parts.push_back({
-                        {"type", part.type},
-                        {"text", part.text},
-                    });
-                }
+                parts.push_back({
+                    {"type", part.type},
+                    {"text", part.text},
+                });
            }
        }
    } else {
@@ -316,7 +309,6 @@ struct common_chat_templates {
    bool has_explicit_template;  // Model had builtin template or template overridden was specified.
    std::unique_ptr<common_chat_template> template_default;  // always set (defaults to chatml)
    std::unique_ptr<common_chat_template> template_tool_use;
-    std::unique_ptr<common_chat_template> template_tts;  // the prompt of a text to speak
 };

 common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::string & tool_choice) {
@@ -378,16 +370,12 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messa
                            throw std::invalid_argument("Missing content part type: " + part.dump());
                        }
                        const auto & type = part.at("type");
-                        common_chat_msg_content_part msg_part;
-                        msg_part.type = type;
-                        if (type == "text" || type == "media_marker") {
-                            msg_part.text = part.at("text");
-                        } else if (type == "tool_reference") {
-                            // names a request tool, the template renders its definition
-                            msg_part.name = part.at("name");
-                        } else {
+                        if (type != "text" && type != "media_marker") {
                            throw std::invalid_argument("Unsupported content part type: " + type.dump());
                        }
+                        common_chat_msg_content_part msg_part;
+                        msg_part.type = type;
+                        msg_part.text = part.at("text");
                        msg.content_parts.push_back(msg_part);
                    }
                } else if (!content.is_null()) {
@@ -440,41 +428,6 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messa
            if (message.contains("tool_call_id")) {
                msg.tool_call_id = message.at("tool_call_id");
            }
-            {
-                json extra = json::object();
-                for (const auto & [k, v] : message.items()) {
-                    if (k != "role" && k != "content" && k != "tool_calls" && k != "reasoning_content" &&
-                        k != "name" && k != "tool_call_id") {
-                        extra[k] = v;
-                    }
-                }
-                if (has_tool_calls) {
-                    // per tool call: keys besides type/function/id, and function keys besides name/arguments
-                    json tc_extra = json::array();
-                    bool any      = false;
-                    for (const auto & tool_call : message.at("tool_calls")) {
-                        json e = json::object();
-                        for (const auto & [k, v] : tool_call.items()) {
-                            if (k != "type" && k != "function" && k != "id") {
-                                e[k] = v;
-                            }
-                        }
-                        for (const auto & [k, v] : tool_call.at("function").items()) {
-                            if (k != "name" && k != "arguments") {
-                                e["function"][k] = v;
-                            }
-                        }
-                        any = any || !e.empty();
-                        tc_extra.push_back(e);
-                    }
-                    if (any) {
-                        extra["__tool_call_extras"] = tc_extra;
-                    }
-                }
-                if (!extra.empty()) {
-                    msg.extra_json = extra.dump();
-                }
-            }

            msgs.push_back(msg);
        }
@@ -563,50 +516,6 @@ static json render_message_to_json(const std::vector<common_chat_msg> & msgs, co
    return messages_inp_normalizer(c).normalize(messages);
 }

-// DeepSeek V4.1 reads non-OpenAI message keys and joins content blocks with "\n\n" around images,
-// so keep media markers as separate text parts and restore the extra keys
-static json render_message_to_json_deepseek_v4_1(const std::vector<common_chat_msg> & msgs, const jinja::caps & c) {
-    json messages = json::array();
-    for (const auto & msg : msgs) {
-        json jmsg;
-        if (msg.contains_media() && msg.content.empty()) {
-            common_chat_msg typed = msg;
-            for (auto & part : typed.content_parts) {
-                part.type = "text";
-            }
-            jmsg = typed.to_json_oaicompat(/* concat_typed_text= */ false);
-        } else {
-            jmsg = msg.to_json_oaicompat(/* concat_typed_text= */ false);
-        }
-        if (!msg.extra_json.empty()) {
-            const json extra = json::parse(msg.extra_json);
-            for (const auto & [k, v] : extra.items()) {
-                if (k == "__tool_call_extras") {
-                    if (!jmsg.contains("tool_calls")) {
-                        continue;
-                    }
-                    auto & tcs = jmsg.at("tool_calls");
-                    for (size_t i = 0; i < tcs.size() && i < v.size(); i++) {
-                        for (const auto & [tk, tv] : v[i].items()) {
-                            if (tk == "function") {
-                                for (const auto & [fk, fv] : tv.items()) {
-                                    tcs[i]["function"][fk] = fv;
-                                }
-                            } else {
-                                tcs[i][tk] = tv;
-                            }
-                        }
-                    }
-                } else if (!jmsg.contains(k)) {
-                    jmsg[k] = v;
-                }
-            }
-        }
-        messages.push_back(std::move(jmsg));
-    }
-    return messages_inp_normalizer(c).normalize(messages);
-}
-
 // DEPRECATED: only used in tests
 json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msgs, bool concat_typed_text) {
    jinja::caps c;
@@ -622,17 +531,13 @@ json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_tool> & t

    auto result = json::array();
    for (const auto & tool : tools) {
-        json function {
-            { "name",        tool.name },
-            { "description", tool.description },
-            { "parameters",  json::parse(tool.parameters) },
-        };
-        if (tool.defer_loading) {
-            function["defer_loading"] = true;
-        }
        result.push_back({
            { "type",     "function" },
-            { "function", function },
+            { "function", {
+                { "name", tool.name },
+                { "description", tool.description },
+                { "parameters", json::parse(tool.parameters) },
+            }},
        });
    }
    return result;
@@ -669,19 +574,10 @@ std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & too
                }

                const auto & function = tool.at("function");
-
-                // "defer_loading" sits on the function, or on the tool next to "type"; the function wins
-                const json & defer = function.contains("defer_loading") ? function.at("defer_loading")
-                                                                        : tool.value("defer_loading", json(false));
-                if (!defer.is_boolean()) {
-                    throw std::invalid_argument("Expected 'defer_loading' to be a boolean, got " + defer.dump());
-                }
-
                result.push_back({
                    /* .name = */ function.at("name"),
                    /* .description = */ function.value("description", ""),
                    /* .parameters = */ function.value("parameters", json::object()).dump(),
-                    /* .defer_loading = */ defer.get<bool>(),
                });
            }
        }
@@ -804,70 +700,19 @@ bool common_chat_templates_was_explicit(const struct common_chat_templates * tmp
    return tmpls->has_explicit_template;
 }

-// renders a task template (e.g. "tts") with the given variables
-static std::string common_chat_template_render(const common_chat_template & tmpl, const json & vars) {
-    jinja::context ctx(tmpl.source());
-    jinja::global_from_json(ctx, vars, false);
-    ctx.set_val("bos_token", jinja::mk_val<jinja::value_string>(tmpl.bos_token()));
-    ctx.set_val("eos_token", jinja::mk_val<jinja::value_string>(tmpl.eos_token()));
-    jinja::runtime runtime(ctx);
-    return jinja::runtime::gather_string_parts(runtime.execute(tmpl.prog))->as_string().str();
-}
-
-// LFM3 format detection: template renders deferred tools inside <additional_tools>...</additional_tools>
-static bool is_lfm3_template(const std::string & src) {
-    return src.find("<additional_tools>") != std::string::npos;
-}
-
-// LFM3-Audio: a known language code maps to its name, no language lets the model detect it
-static std::string lfm3_asr_system_prompt(const std::string & language) {
-    static const std::map<std::string, std::string> names = {
-        { "ar", "Arabic" }, { "da", "Danish" }, { "de", "German" }, { "el", "Greek" }, { "en", "English" },
-        { "es", "Spanish" }, { "fi", "Finnish" }, { "fr", "French" }, { "he", "Hebrew" }, { "it", "Italian" },
-        { "ja", "Japanese" }, { "ko", "Korean" }, { "nl", "Dutch" }, { "pl", "Polish" }, { "pt", "Portuguese" },
-        { "sv", "Swedish" }, { "tr", "Turkish" }, { "zh", "Chinese" },
-    };
-    if (language.empty()) {
-        return "Perform ASR.";
-    }
-    std::string code = language;
-    std::transform(code.begin(), code.end(), code.begin(), [](unsigned char c) { return (char) std::tolower(c); });
-    const auto it = names.find(code);
-    return "Perform ASR in " + (it != names.end() ? it->second : language) + ".";
-}
-
-common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates, const std::string & language) {
+common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates) {
    common_chat_prompt_preset asr_preset;
    asr_preset.system = "";
    asr_preset.user   = "Transcribe audio to text";

-    if (chat_templates && chat_templates->template_default) {
-        const std::string src = chat_templates->template_default->source();
-        if (is_lfm2_template(src)) {
-            asr_preset.system = "Perform ASR.";
-            asr_preset.user   = "";
-        } else if (is_lfm3_template(src)) {
-            asr_preset.system = lfm3_asr_system_prompt(language);
-            asr_preset.user   = "";
-            asr_preset.has_language = true;
-        }
+    if (chat_templates && chat_templates->template_default && is_lfm2_template(chat_templates->template_default->source())) {
+        asr_preset.system = "Perform ASR.";
+        asr_preset.user   = "";
    }

    return asr_preset;
 }

-std::string common_chat_get_tts_prompt(const common_chat_templates * chat_templates, const std::string & text,
-                                       const std::string & language, const std::string & voice) {
-    if (!chat_templates || !chat_templates->template_tts) {
-        return "";
-    }
-    return common_chat_template_render(*chat_templates->template_tts, {
-        {"messages", json::array({{{"role", "user"}, {"content", text}}})},
-        {"language", language},
-        {"voice",    voice},
-    });
-}
-
 std::string common_chat_templates_source(const struct common_chat_templates * tmpls, const std::string & variant) {
    if (!variant.empty()) {
        if (variant == "tool_use") {
@@ -978,16 +823,6 @@ common_chat_templates_ptr common_chat_templates_init(const struct llama_model *
            LOG_ERR("%s: failed to parse tool use chat template (ignoring it): %s\n", __func__, e.what());
        }
    }
-    // task templates of the model, kept with a chat template override
-    if (model) {
-        if (const char * src = llama_model_chat_template(model, "tts")) {
-            try {
-                tmpls->template_tts = std::make_unique<common_chat_template>(src, token_bos, token_eos);
-            } catch (const std::exception & e) {
-                LOG_ERR("%s: failed to parse the tts chat template (ignoring it): %s\n", __func__, e.what());
-            }
-        }
-    }
    return tmpls;
 }

@@ -1373,13 +1208,11 @@ static common_chat_params_init_fn common_chat_template_detect_params_init(const
    // DeepSeek V3.2/V4 format detection: template defines dsml_token and uses it for tool calls.
    // The template source contains the token as a variable assignment, not as a literal in markup.
    // V3.2 names the tool call block "function_calls", V4 names it "tool_calls".
-    // V4.1 tag names have a leading space and its template does not name the dsml_token variable.
-    if ((src.find("dsml_token") != std::string::npos &&
-         src.find("DSML") != std::string::npos &&
-         (src.find("function_calls") != std::string::npos ||
-          src.find("tool_calls") != std::string::npos)) ||
-        src.find("｜DSML｜ calls>") != std::string::npos) {
-        LOG_DBG("Using specialized template: DeepSeek V3.2/V4/V4.1\n");
+    if (src.find("dsml_token") != std::string::npos &&
+        src.find("DSML") != std::string::npos &&
+        (src.find("function_calls") != std::string::npos ||
+         src.find("tool_calls") != std::string::npos)) {
+        LOG_DBG("Using specialized template: DeepSeek V3.2/V4\n");
        return common_chat_params_init_deepseek_v3_2;
    }

@@ -1480,9 +1313,7 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
        workaround::trim_all_content(trimmed_messages);
        messages_to_render = &trimmed_messages;
    }
-    params.messages              = src.find("｜DSML｜ calls>") != std::string::npos ?
-                                       render_message_to_json_deepseek_v4_1(*messages_to_render, tmpl.original_caps()) :
-                                       render_message_to_json(*messages_to_render, tmpl.original_caps());
+    params.messages              = render_message_to_json(*messages_to_render, tmpl.original_caps());
    params.tool_choice           = inputs.tool_choice;
    params.reasoning_format      = inputs.reasoning_format;
    params.enable_thinking       = inputs.enable_thinking;
@@ -1803,9 +1634,6 @@ common_chat_msg common_chat_peg_parse(const common_peg_arena &          src_pars
    //LOG_DBG("Parsing PEG input with format %s: %s\n", common_chat_format_name(params.format), effective_input.text.c_str());

    common_peg_parse_flags flags = COMMON_PEG_PARSE_FLAG_LENIENT;
-    if (!is_partial) {
-        flags |= COMMON_PEG_PARSE_FLAG_COMPLETE;
-    }
    if (params.debug) {
        flags |= COMMON_PEG_PARSE_FLAG_DEBUG;
    }
@@ -1943,13 +1771,12 @@ const common_chat_msg & common_chat_session::finish(const common_chat_input & ch
    return result;
 }

-common_json common_chat_templates_get_caps(const common_chat_templates * chat_templates) {
+std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_templates * chat_templates) {
    GGML_ASSERT(chat_templates != nullptr);
    GGML_ASSERT(chat_templates->template_default != nullptr);
-    // take the more expressive template when available
-    const auto & caps = chat_templates->template_tool_use != nullptr ? chat_templates->template_tool_use->caps
-                                                                      : chat_templates->template_default->caps;
-    common_json result = caps.to_map();
-    result["reasoning_effort_levels"] = caps.reasoning_effort_levels;
-    return result;
+    if (chat_templates->template_tool_use != nullptr) {
+        // take the more expressive template when available
+        return chat_templates->template_tool_use->caps.to_map();
+    }
+    return chat_templates->template_default->caps.to_map();
 }
@@ -45,7 +45,6 @@ struct common_chat_tool_call {
 struct common_chat_msg_content_part {
    std::string type;
    std::string text;
-    std::string name; // tool_reference: name of a tool from the request tools

    // TODO @ngxson : no known chat templates support reasoning_content in content parts yet
    //                this can be useful for models with interleaved thinking (like Kimi-K2)
@@ -53,7 +52,7 @@ struct common_chat_msg_content_part {
    // std::string reasoning_content;

    bool operator==(const common_chat_msg_content_part & other) const {
-        return type == other.type && text == other.text && name == other.name;
+        return type == other.type && text == other.text;
    }
 };

@@ -95,8 +94,6 @@ struct common_chat_msg {
    std::string                               reasoning_content;
    std::string                               tool_name;
    std::string                               tool_call_id;
-    // non-OpenAI message keys as a JSON object string, only for templates that read them (DeepSeek V4.1)
-    std::string                               extra_json;

    common_json to_json_oaicompat(bool concat_typed_text = false) const;

@@ -227,9 +224,6 @@ struct common_chat_tool {
    std::string name;
    std::string description;
    std::string parameters;
-
-    // withheld from the system prompt, a tool_reference reveals it later in the conversation
-    bool defer_loading = false;
 };

 enum common_chat_tool_choice {
@@ -472,7 +466,7 @@ common_json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_to
 common_json common_chat_tool_parameters(const common_json & function);

 // get template caps, useful for reporting to server /props endpoint
-common_json common_chat_templates_get_caps(const common_chat_templates * chat_templates);
+std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_templates * chat_templates);

 common_chat_prompt common_chat_template_direct_apply(
    const common_chat_template & tmpl,
@@ -491,12 +485,6 @@ std::optional<common_chat_params> common_chat_try_specialized_template(
 struct common_chat_prompt_preset {
    std::string system;
    std::string user;
-    bool has_language = false; // the preset already names the requested language
 };

-common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates, const std::string & language = "");
-
-// prompt of the model's "tts" chat template for the text to speak, empty if the model has none
-// language and voice are optional, the template defines their defaults
-std::string common_chat_get_tts_prompt(const common_chat_templates * chat_templates, const std::string & text,
-                                       const std::string & language = "", const std::string & voice = "");
+common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates);
@@ -26,7 +26,6 @@
 #include <fstream>
 #include <iostream>
 #include <iterator>
-#include <random>
 #include <regex>
 #include <sstream>
 #include <string>
@@ -219,23 +218,6 @@ int32_t common_cpu_get_num_math() {
            }
        }
    }
-#elif defined(__riscv) && defined(__linux__)
-    std::ifstream cpuinfo("/proc/cpuinfo");
-    std::string line;
-    int32_t n_a100 = 0;
-    while (std::getline(cpuinfo, line)) {
-        std::istringstream fields(line);
-        std::string key;
-        std::string separator;
-        uint64_t marchid;
-        if (fields >> key >> separator >> std::hex >> marchid &&
-            key == "marchid" && separator == ":" && marchid == UINT64_C(0x8000000041000002)) {
-            ++n_a100;
-        }
-    }
-    if (n_a100 > 0) {
-        return n_a100;
-    }
 #elif defined(__powerpc64__) || defined(__powerpc__)
    int32_t smt_factor = 1;
    int phy_cpus = common_cpu_get_num_physical_cores();
@@ -759,56 +741,50 @@ static inline bool glob_class_match(const char c, const char * pattern, const ch
 }

 // simple glob: * matches non-/ chars, ** matches anything including /, [] matches character class
-// only wildcards recurse, too many wildcards: no match
-static bool glob_match(const char * pattern, const char * str, int depth = 0) {
-    if (depth >= COMMON_RECURSIVE_LIMIT) {
+static inline bool glob_match(const char * pattern, const char * str) {
+    if (*pattern == '\0') {
+        return *str == '\0';
+    }
+    if (pattern[0] == '*' && pattern[1] == '*') {
+        const char * p = pattern + 2;
+        if (glob_match(p, str)) return true;
+        if (*str != '\0') return glob_match(pattern, str + 1);
        return false;
    }
-    while (*pattern != '\0') {
-        if (pattern[0] == '*' && pattern[1] == '*') {
-            const char * p = pattern + 2;
-            for (;; str++) {
-                if (glob_match(p, str, depth + 1)) return true;
-                if (*str == '\0') return false;
-            }
+    if (*pattern == '*') {
+        const char * p = pattern + 1;
+        for (; *str != '\0' && *str != '/'; str++) {
+            if (glob_match(p, str)) return true;
        }
-        if (*pattern == '*') {
-            const char * p = pattern + 1;
-            for (; *str != '\0' && *str != '/'; str++) {
-                if (glob_match(p, str, depth + 1)) return true;
-            }
-            return glob_match(p, str, depth + 1);
-        }
-        if (*pattern == '?' && *str != '\0' && *str != '/') {
-            pattern++;
-            str++;
-            continue;
-        }
-        if (*pattern == '[') {
-            const char * class_end = pattern + 1;
-            // If first character after '[' is ']' or '-', treat it as literal
-            if (*class_end == ']' || *class_end == '-') {
-                class_end++;
-            }
-            while (*class_end != '\0' && *class_end != ']') {
-                class_end++;
-            }
-            if (*class_end == ']') {
-                if (*str == '\0' || !glob_class_match(*str, pattern + 1, class_end)) return false;
-                pattern = class_end + 1;
-                str++;
-                continue;
-            }
-            if (*str != '[') return false;
-            pattern++;
-            str++;
-            continue;
-        }
-        if (*pattern != *str) return false;
-        pattern++;
-        str++;
+        return glob_match(p, str);
    }
-    return *str == '\0';
+    if (*pattern == '?' && *str != '\0' && *str != '/') {
+        return glob_match(pattern + 1, str + 1);
+    }
+    if (*pattern == '[') {
+        const char * class_end = pattern + 1;
+        // If first character after '[' is ']' or '-', treat it as literal
+        if (*class_end == ']' || *class_end == '-') {
+            class_end++;
+        }
+        while (*class_end != '\0' && *class_end != ']') {
+            class_end++;
+        }
+        if (*class_end == ']') {
+            if (*str == '\0') return false;
+            bool matched = glob_class_match(*str, pattern + 1, class_end);
+            return matched && glob_match(class_end + 1, str + 1);
+        } else {
+            if (*str == '[') {
+                return glob_match(pattern + 1, str + 1);
+            }
+            return false;
+        }
+    }
+    if (*pattern == *str) {
+        return glob_match(pattern + 1, str + 1);
+    }
+    return false;
 }

 bool glob_match(const std::string & pattern, const std::string & str) {
@@ -934,14 +910,16 @@ std::string wstring_to_utf8(const std::wstring & str) {
 }
 #endif

-void fs_write_atomic(const std::filesystem::path & path, const std::string & data) {
-    // a random suffix gives every writer its own temporary file, across processes and containers
-    std::random_device rd;
-    const uint64_t id = ((uint64_t) rd() << 32) | rd();
+// returns the path as a UTF-8 string, preserving its separators
+std::string fs_path_to_utf8(const std::filesystem::path & path) {
+    const auto value = path.u8string();
+    return std::string(value.begin(), value.end());
+}

+void fs_write_atomic(const std::filesystem::path & path, const std::string & data) {
    std::error_code ec;
    std::filesystem::path path_tmp = path;
-    path_tmp += "." + std::to_string(id) + ".tmp";
+    path_tmp += ".tmp";

    if (path.has_parent_path()) {
        std::filesystem::create_directories(path.parent_path(), ec);
@@ -957,7 +935,7 @@ void fs_write_atomic(const std::filesystem::path & path, const std::string & dat

    if (file.fail() || ec) {
        std::filesystem::remove(path_tmp, ec);
-        throw std::runtime_error("failed to write file: " + path.string());
+        throw std::runtime_error("failed to write file: " + fs_path_to_utf8(path));
    }
 }

@@ -1051,7 +1029,7 @@ std::filesystem::path fs_get_cache_file(const std::string & filename) {
    std::error_code ec;
    common_create_directories(cache_directory, ec);
    if (ec) {
-        throw std::runtime_error("failed to create cache directory: " + cache_directory.string());
+        throw std::runtime_error("failed to create cache directory: " + fs_path_to_utf8(cache_directory));
    }
    return cache_directory / std::filesystem::u8path(filename);
 }
@@ -1198,9 +1176,6 @@ static const std::map<common_decision_type, std::string> COMMON_DECISION_TYPE_NA
    { COMMON_DECISION_TYPE_LFM2_D1_OMNI,   "lfm2-d1-omni"  },
    { COMMON_DECISION_TYPE_GRANITE_SWITCH, "granite-switch"},
    { COMMON_DECISION_TYPE_H2O_LIGHTNING,  "h2o-lightning" },
-    { COMMON_DECISION_TYPE_MINICPM_OCR_JEV, "openbmb-minicpm-ocr-jev" },
-    { COMMON_DECISION_TYPE_LFM3_D1,        "lfm3-d1"       },
-    { COMMON_DECISION_TYPE_LFM3_D1_OMNI,   "lfm3-d1-omni"  },
    // for future models, ALWAYS add org name prefix to avoid name collisions, e.g. "orgname-modelname"
 };

@@ -1266,50 +1241,44 @@ common_gguf_info common_get_gguf_info(const std::string & fname) {
    return info;
 }

-void common_fit_apply(common_params & params, llama_model_params & mparams, llama_context_params & cparams) {
-    if (!params.fit_params) {
-        return;
-    }
-
-    COM_TRC("%s", "fitting params to device memory ...\n");
-    COM_TRC("%s", "(for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)\n");
-
-    // the draft context is created from the same base params and follows the main context, fit both together
-    const bool has_draft = params.speculative.has_dft();
-    const bool spec_mtp  = std::find(params.speculative.types.begin(), params.speculative.types.end(),
-        COMMON_SPECULATIVE_TYPE_DRAFT_MTP) != params.speculative.types.end();
-
-    common_params params_dft = common_base_params_to_speculative(params);
-
-    auto mparams_dft = common_model_params_to_llama(params_dft);
-    auto cparams_dft = common_context_params_to_llama(params_dft);
-    if (spec_mtp) {
-        cparams_dft.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
-    }
-    cparams_dft.n_rs_seq = 0;
-
-    const common_fit_extra_model extra = {
-        /*.path_model   =*/ params_dft.model.path.c_str(),
-        /*.mparams      =*/ &mparams_dft,
-        /*.cparams      =*/ &cparams_dft,
-        /*.shares_model =*/ !has_draft, // an MTP context runs on the weights of the main model
-    };
-
-    common_fit_params(params.model.path.c_str(), &mparams, &cparams,
-        params.tensor_split,
-        params.tensor_buft_overrides.data(),
-        params.fit_params_target.data(),
-        params.fit_params_min_ctx,
-        has_draft || spec_mtp ? &extra : nullptr,
-        params.verbosity >= LOG_LEVEL_DEBUG ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR);
-}
-
 common_init_result::common_init_result(common_params & params, bool model_only) :
    pimpl(new impl{}) {
    auto mparams = common_model_params_to_llama(params);
    auto cparams = common_context_params_to_llama(params);

-    common_fit_apply(params, mparams, cparams);
+    if (params.fit_params) {
+        COM_TRC("%s", "fitting params to device memory ...\n");
+        COM_TRC("%s", "(for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)\n");
+
+        // the draft context is created from the same base params and follows the main context, fit both together
+        const bool has_draft = params.speculative.has_dft();
+        const bool spec_mtp  = std::find(params.speculative.types.begin(), params.speculative.types.end(),
+            COMMON_SPECULATIVE_TYPE_DRAFT_MTP) != params.speculative.types.end();
+
+        common_params params_dft = common_base_params_to_speculative(params);
+
+        auto mparams_dft = common_model_params_to_llama(params_dft);
+        auto cparams_dft = common_context_params_to_llama(params_dft);
+        if (spec_mtp) {
+            cparams_dft.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
+        }
+        cparams_dft.n_rs_seq = 0;
+
+        const common_fit_extra_model extra = {
+            /*.path_model   =*/ params_dft.model.path.c_str(),
+            /*.mparams      =*/ &mparams_dft,
+            /*.cparams      =*/ &cparams_dft,
+            /*.shares_model =*/ !has_draft, // an MTP context runs on the weights of the main model
+        };
+
+        common_fit_params(params.model.path.c_str(), &mparams, &cparams,
+            params.tensor_split,
+            params.tensor_buft_overrides.data(),
+            params.fit_params_target.data(),
+            params.fit_params_min_ctx,
+            has_draft || spec_mtp ? &extra : nullptr,
+            params.verbosity >= LOG_LEVEL_DEBUG ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR);
+    }

    llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams);
    if (model == NULL) {
@@ -1327,17 +1296,13 @@ common_init_result::common_init_result(common_params & params, bool model_only)
    // these decision models return a score for each token via the embeddings output
    // TODO: maybe improve this in the future
    const auto decision_type = common_get_decision_type(model);
-    // minicpm-ocr-jev reads its class scores from the pooled output, the pooling type is the one of the GGUF
    if (decision_type == COMMON_DECISION_TYPE_LAYA || decision_type == COMMON_DECISION_TYPE_KEV || decision_type == COMMON_DECISION_TYPE_CLEF ||
-        decision_type == COMMON_DECISION_TYPE_LFM2_D1_OMNI || decision_type == COMMON_DECISION_TYPE_VLLM_DECISION2 ||
-        decision_type == COMMON_DECISION_TYPE_MINICPM_OCR_JEV) {
-        const auto pooling_type = decision_type == COMMON_DECISION_TYPE_MINICPM_OCR_JEV ? LLAMA_POOLING_TYPE_UNSPECIFIED : LLAMA_POOLING_TYPE_NONE;
-
+        decision_type == COMMON_DECISION_TYPE_LFM2_D1_OMNI || decision_type == COMMON_DECISION_TYPE_VLLM_DECISION2) {
        params.embedding    = true;
-        params.pooling_type = pooling_type;
+        params.pooling_type = LLAMA_POOLING_TYPE_NONE;

        cparams.embeddings            = true;
-        cparams.pooling_type          = pooling_type;
+        cparams.pooling_type          = LLAMA_POOLING_TYPE_NONE;
        cparams.n_outputs_max         = cparams.n_batch;
        cparams.n_outputs_max_per_seq = 1;

@@ -1484,7 +1449,7 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
        if (params.control_vector_layer_start <= 0) params.control_vector_layer_start = 1;
        if (params.control_vector_layer_end   <= 0) params.control_vector_layer_end   = llama_model_n_layer(model);

-        const auto cvec = common_control_vector_load(params.control_vectors, llama_model_n_layer(model));
+        const auto cvec = common_control_vector_load(params.control_vectors);
        if (cvec.n_embd == -1) {
            return res;
        }
@@ -2006,7 +1971,7 @@ float common_embd_similarity_cos(const float * embd1, const float * embd2, int n
 // Control vector utils
 //

-static common_control_vector_data common_control_vector_load_one(const common_control_vector_load_info & load_info, int32_t n_layer) {
+static common_control_vector_data common_control_vector_load_one(const common_control_vector_load_info & load_info) {
    common_control_vector_data result = { -1, {} };

    ggml_context * ctx = nullptr;
@@ -2047,10 +2012,6 @@ static common_control_vector_data common_control_vector_load_one(const common_co
            COM_ERR("invalid (zero) direction tensor layer index in %s\n", load_info.fname.c_str());
            result.n_embd = -1;
            break;
-        } else if (layer_idx > n_layer) {
-            COM_ERR("direction tensor layer index %d exceeds the model layer count %d in %s\n", layer_idx, n_layer, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
        }

        struct ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());
@@ -2074,10 +2035,10 @@ static common_control_vector_data common_control_vector_load_one(const common_co
        }

        // extend if necessary - do not store data for layer 0 (it's not used)
-        result.data.resize(std::max(result.data.size(), (size_t) result.n_embd * layer_idx), 0.0f);
+        result.data.resize(std::max(result.data.size(), static_cast<size_t>(result.n_embd * layer_idx)), 0.0f);

        const float * src = (const float *) tensor->data;
-        float * dst = result.data.data() + (size_t) result.n_embd * (layer_idx - 1);  // layer 1 at [0]
+        float * dst = result.data.data() + result.n_embd * (layer_idx - 1);  // layer 1 at [0]
        for (int j = 0; j < result.n_embd; j++) {
            dst[j] += src[j] * load_info.strength;  // allows multiple directions for same layer in same file
        }
@@ -2095,11 +2056,11 @@ static common_control_vector_data common_control_vector_load_one(const common_co
    return result;
 }

-common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos, int32_t n_layer) {
+common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos) {
    common_control_vector_data result = { -1, {} };

    for (const auto & info : load_infos) {
-        auto cur = common_control_vector_load_one(info, n_layer);
+        auto cur = common_control_vector_load_one(info);

        if (cur.n_embd == -1) {
            result.n_embd = -1;
@@ -31,8 +31,6 @@
 #define DIRECTORY_SEPARATOR '/'
 #endif // _WIN32

-#define COMMON_RECURSIVE_LIMIT 128
-
 #define COM_DBG(fmt, ...) LOG_DBG("cmn  %12.*s: " fmt, 12, __func__, __VA_ARGS__)
 #define COM_TRC(fmt, ...) LOG_TRC("cmn  %12.*s: " fmt, 12, __func__, __VA_ARGS__)
 #define COM_INF(fmt, ...) LOG_INF("cmn  %12.*s: " fmt, 12, __func__, __VA_ARGS__)
@@ -701,7 +699,6 @@ struct common_params {
    std::string models_dir    = "";     // directory containing models for the router server
    std::string models_preset = "";     // directory containing model presets for the router server
    int models_max = 4;                 // maximum number of models to load simultaneously
-    int models_memory_margin = 0;       // MiB of device memory to keep free when loading models (0 = no memory accounting)
    bool models_autoload = true;        // automatically load models when requested via the router server
    std::string models_preset_hf = "";  // show a warning about remote presets on router loaded (if not empty)

@@ -772,7 +769,6 @@ struct common_params {

    // TTS params
    std::string tts_lang = "";
-    std::string tts_voice = "";
    std::string tts_speaker_file = "";

    bool is_gen_docs = false; // whether we are running inside llama-gen-docs
@@ -910,6 +906,9 @@ std::wstring utf8_to_wstring(const std::string & str);
 std::string  wstring_to_utf8(const std::wstring & str);
 #endif

+// returns the path as a UTF-8 string, preserving its separators
+std::string fs_path_to_utf8(const std::filesystem::path & path);
+
 //
 // Environment utils
 //
@@ -977,9 +976,6 @@ enum common_decision_type {
    COMMON_DECISION_TYPE_LFM2_D1_OMNI, // same as laya, other prompt layout
    COMMON_DECISION_TYPE_GRANITE_SWITCH, // same as openjev, run by the decision aLoRA of granite switch
    COMMON_DECISION_TYPE_H2O_LIGHTNING, // same as openjev, the labels are read after an answer prefill
-    COMMON_DECISION_TYPE_MINICPM_OCR_JEV, // one score per class, the pooled embeddings output of the last token
-    COMMON_DECISION_TYPE_LFM3_D1,       // same as lfm2-d1, non-causal, label codes A..Z then AA, AB..
-    COMMON_DECISION_TYPE_LFM3_D1_OMNI,  // same as lfm3-d1
    COMMON_DECISION_TYPE_UNKNOWN, // a decision model of a type that is not supported
 };

@@ -1019,9 +1015,6 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
 struct llama_model_params   common_model_params_to_llama  (      common_params & params);
 struct llama_context_params common_context_params_to_llama(const common_params & params);

-// apply --fit to mparams and cparams, no-op if params.fit_params is false
-void common_fit_apply(common_params & params, struct llama_model_params & mparams, struct llama_context_params & cparams);
-
 // clear LoRA adapters from context, then apply new list of adapters
 void common_set_adapter_lora(struct llama_context * ctx, std::vector<common_adapter_lora_info> & lora);

@@ -1234,9 +1227,8 @@ struct common_control_vector_load_info {
 };

 // Load control vectors, scale each by strength, and add them together.
-// Direction layer indices must be in [1, n_layer].
 // On error, returns {-1, empty}
-common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos, int32_t n_layer);
+common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos);

 //
 // Split utils
@@ -4,7 +4,6 @@
 #include "common.h"
 #include "log.h"
 #include "download.h"
-#include "file-lock.h"
 #include "hf-cache.h"
 #include "json.h"

@@ -78,18 +77,6 @@ std::pair<std::string, std::string> common_download_split_repo_tag(const std::st
    return {hf_repo, tag};
 }

-// file name from url, without the query string
-static std::string url_get_filename(const std::string & url) {
-    std::string filename = url;
-    if (auto pos = filename.rfind('/'); pos != std::string::npos) {
-        filename = filename.substr(pos + 1);
-    }
-    if (auto pos = filename.find('?'); pos != std::string::npos) {
-        filename = filename.substr(0, pos);
-    }
-    return filename;
-}
-
 class ProgressBar : public common_download_callback {
    static inline std::mutex mutex;
    static inline std::map<const ProgressBar *, int> lines;
@@ -113,8 +100,14 @@ public:
    ProgressBar() = default;

    void on_start(const common_download_progress & p) override {
-        filename = url_get_filename(p.url);
+        filename = p.url;

+        if (auto pos = filename.rfind('/'); pos != std::string::npos) {
+            filename = filename.substr(pos + 1);
+        }
+        if (auto pos = filename.find('?'); pos != std::string::npos) {
+            filename = filename.substr(0, pos);
+        }
        for (size_t i = 0; i < filename.size(); ++i) {
            if ((filename[i] & 0xC0) != 0x80) {
                if (len++ == 39) {
@@ -264,62 +257,15 @@ static bool common_pull_file(httplib::Client & cli,
 static int common_download_file_single_online(const std::string & url,
                                              const std::string & path,
                                              const common_download_opts & opts,
-                                              bool skip_etag,
-                                              const std::string & lock_path,
-                                              const std::string & alt_path) {
+                                              bool skip_etag) {
    static const int max_attempts        = 3;
    static const int retry_delay_seconds = 2;

-    const std::string path_temporary = path + PARTIAL_SUFFIX;
-    const std::string path_lock = lock_path.empty() ? path + ".lock" : lock_path;
-
-    common_download_progress p;
-    p.url = url;
-
-    if (opts.callback) {
-        opts.callback->on_start(p);
-    }
-
-    // the completion callback may itself take the same lock (the HF cache
-    // finalizes blobs under it), so every exit goes through finish()
-    common_file_lock lock(path_lock);
-    auto finish = [&](int status) {
-        lock.close();
-        if (opts.callback) {
-            opts.callback->on_done(p, is_http_status_ok(status));
-        }
-        return status;
-    };
-
-    // alt_path is the HF snapshot: the blob may have been moved there when symlinks are unavailable
-    auto cache_hit = [&] {
-        return std::filesystem::exists(std::filesystem::u8path(path)) ||
-               (!alt_path.empty() && std::filesystem::exists(std::filesystem::u8path(alt_path)));
-    };
-
-    if (skip_etag && cache_hit()) {
-        LOG_DBG("%s: using cached file: %s\n", __func__, path.c_str());
-        return finish(304); // 304 Not Modified - fake cached response
-    }
-
-    // take the lock before touching the cache, so two processes cannot race on the same file
-    bool waited = false;
-    if (!lock.acquire([&] {
-            if (!waited) {
-                LOG_INF("%s: file '%s' is being downloaded by another process, waiting...\n", __func__, url_get_filename(url).c_str());
-                waited = true;
-            }
-            return !(opts.callback && opts.callback->is_cancelled());
-        })) {
-        return finish(-1);
-    }
-
-    // under the lock: another process may have finished while we waited
-    bool file_exists = cache_hit();
+    const bool file_exists = std::filesystem::exists(std::filesystem::u8path(path));

    if (file_exists && skip_etag) {
        LOG_DBG("%s: using cached file: %s\n", __func__, path.c_str());
-        return finish(304); // 304 Not Modified - fake cached response
+        return 304; // 304 Not Modified - fake cached response
    }

    auto [cli, parts] = common_http_client(url);
@@ -348,9 +294,9 @@ static int common_download_file_single_online(const std::string & url,
        LOG_TRC("%s: HEAD failed, status: %d\n", __func__, head ? head->status : -1);
        if (file_exists) {
            LOG_TRC("%s: using cached file (HEAD failed): %s\n", __func__, path.c_str());
-            return finish(304); // 304 Not Modified - fake cached response
+            return 304; // 304 Not Modified - fake cached response
        }
-        return finish(head ? head->status : -1);
+        return head ? head->status : -1;
    }

    std::string etag;
@@ -358,6 +304,8 @@ static int common_download_file_single_online(const std::string & url,
        etag = head->get_header_value("ETag");
    }

+    common_download_progress p;
+    p.url = url;
    if (head->has_header("Content-Length")) {
        try {
            p.total = std::stoull(head->get_header_value("Content-Length"));
@@ -374,16 +322,16 @@ static int common_download_file_single_online(const std::string & url,
    if (file_exists) {
        if (etag.empty()) {
            LOG_DBG("%s: using cached file (no server etag): %s\n", __func__, path.c_str());
-            return finish(304); // 304 Not Modified - fake cached response
+            return 304; // 304 Not Modified - fake cached response
        }
        if (!last_etag.empty() && last_etag == etag) {
            LOG_DBG("%s: using cached file (same etag): %s\n", __func__, path.c_str());
-            return finish(304); // 304 Not Modified - fake cached response
+            return 304; // 304 Not Modified - fake cached response
        }
        // pass this point, the file exists but is different from the server version, so we need to redownload it
        if (remove(path.c_str()) != 0) {
            LOG_ERR("%s: unable to delete file: %s\n", __func__, path.c_str());
-            return finish(-1);
+            return -1;
        }
    }

@@ -393,6 +341,7 @@ static int common_download_file_single_online(const std::string & url,
    }

    bool success = false;
+    const std::string path_temporary = path + PARTIAL_SUFFIX;
    int delay = retry_delay_seconds;

    if (!skip_etag && !etag.empty()) {
@@ -400,7 +349,7 @@ static int common_download_file_single_online(const std::string & url,
            LOG_WRN("%s: partial file does not match the server etag, discarding it: %s\n", __func__, path_temporary.c_str());
            if (remove(path_temporary.c_str()) != 0) {
                LOG_ERR("%s: unable to delete file: %s\n", __func__, path_temporary.c_str());
-                return finish(-1);
+                return -1;
            }
        }
        write_etag(path_temporary, etag);
@@ -410,7 +359,7 @@ static int common_download_file_single_online(const std::string & url,
        // report the resume offset, so a resumed file does not start at 0
        p.downloaded = supports_ranges && std::filesystem::exists(path_temporary) ?
                       std::filesystem::file_size(path_temporary) : 0;
-        opts.callback->on_update(p);
+        opts.callback->on_start(p);
    }

    for (int i = 0; i < max_attempts; ++i) {
@@ -454,12 +403,16 @@ static int common_download_file_single_online(const std::string & url,
        }
    }

+    if (opts.callback) {
+        opts.callback->on_done(p, success);
+    }
    // a stopped download keeps path_temporary, so the next attempt resumes from it
    if (!success) {
        LOG_ERR("%s: download failed after %d attempts\n", __func__, max_attempts);
+        return -1; // max attempts reached
    }

-    return finish(success ? head->status : -1);
+    return head->status;
 }

 std::pair<long, std::vector<char>> common_remote_get_content(const std::string          & url,
@@ -499,16 +452,14 @@ std::pair<long, std::vector<char>> common_remote_get_content(const std::string
 int common_download_file_single(const std::string & url,
                                const std::string & path,
                                const common_download_opts & opts,
-                                bool skip_etag,
-                                const std::string & lock_path,
-                                const std::string & alt_path) {
+                                bool skip_etag) {
    if (!opts.offline) {
        ProgressBar tty_cb;
        common_download_opts online_opts = opts;
        if (!online_opts.callback) {
            online_opts.callback = &tty_cb;
        }
-        return common_download_file_single_online(url, path, online_opts, skip_etag, lock_path, alt_path);
+        return common_download_file_single_online(url, path, online_opts, skip_etag);
    }

    if (!std::filesystem::exists(std::filesystem::u8path(path))) {
@@ -870,7 +821,7 @@ void common_download_run_tasks(const std::vector<common_download_task> & tasks)
    for (const auto & task : tasks) {
        futures.push_back(std::async(std::launch::async,
            [&task]() {
-                return common_download_file_single(task.url, task.local_path, task.opts, task.is_hf, task.lock_path, task.alt_path);
+                return common_download_file_single(task.url, task.local_path, task.opts, task.is_hf);
            }
        ));
    }
@@ -1005,7 +956,7 @@ std::string common_docker_resolve_model(const std::string & docker) {
        std::string model_filename = repo;
        std::replace(model_filename.begin(), model_filename.end(), '/', '_');
        model_filename += "_" + tag + ".gguf";
-        std::string local_path = fs_get_cache_file(model_filename).string();
+        std::string local_path = fs_path_to_utf8(fs_get_cache_file(model_filename));

        const std::string blob_url = url_prefix + "/blobs/" + gguf_digest;
        common_download_opts opts;
@@ -69,15 +69,12 @@ struct common_download_task {
    std::string local_path;
    std::function<void()> on_done;
    bool is_hf = false;
-    std::string lock_path;
-    std::string alt_path; // HF snapshot path, also treated as cached (the blob may be moved there)

    common_download_task() = default;
    common_download_task(hf_cache::hf_file f,
            const common_download_opts & opts,
            std::function<void()> on_done = nullptr)
-        : opts(opts), url(f.url), local_path(f.local_path), on_done(on_done), is_hf(true),
-          lock_path(hf_cache::get_lock_path(f.repo_id, f.oid)), alt_path(f.final_path) {}
+        : opts(opts), url(f.url), local_path(f.local_path), on_done(on_done), is_hf(true) {}
 };

 void common_download_run_tasks(const std::vector<common_download_task> & tasks);
@@ -95,14 +92,10 @@ std::string common_download_resolve_path(const std::string & hf_repo_with_tag, c
 // download single file from url to local path
 // returns status code or -1 on error
 // skip_etag: if true, don't read/write .etag files (for HF cache where filename is the hash)
-// lock_path: lock file used to serialize downloads of the same file; when empty, "<path>.lock" is used
-// alt_path: additional path treated as present in cache (see cache_hit in download.cpp)
 int common_download_file_single(const std::string & url,
                                const std::string & path,
                                const common_download_opts & opts = {},
-                                bool skip_etag = false,
-                                const std::string & lock_path = "",
-                                const std::string & alt_path = "");
+                                bool skip_etag = false);

 // resolve and download model from Docker registry
 // return local path to downloaded model file
@@ -1,179 +0,0 @@
-#include "file-lock.h"
-
-#include "log.h"
-
-#include <chrono>
-#include <filesystem>
-#include <thread>
-
-#include <fcntl.h>
-#include <sys/stat.h>
-
-#if defined(_WIN32)
-#define WIN32_LEAN_AND_MEAN
-#ifndef NOMINMAX
-#define NOMINMAX
-#endif
-#include <windows.h>
-#include <io.h>
-#else
-#include <sys/file.h>
-#include <unistd.h>
-#include <cerrno>
-#include <cstring>
-#endif
-
-common_file_lock::common_file_lock(const std::string & lock_path) {
-    this->lock_path = lock_path;
-}
-
-#if defined(_WIN32)
-// the \\?\ prefix (for paths longer than 255 chars) is only accepted on absolute paths
-// with backslash separators: normalize first, then check the length of the absolute form
-static std::wstring as_extended_path(const std::string & path) {
-    // convert from the narrow encoding std::filesystem uses on this system
-    int wlen = MultiByteToWideChar(CP_ACP, 0, path.c_str(), (int) path.size(), nullptr, 0);
-    if (wlen <= 0) {
-        return std::wstring(path.begin(), path.end()); // best effort for exotic encodings
-    }
-    std::wstring wide(wlen, L'\0');
-    MultiByteToWideChar(CP_ACP, 0, path.c_str(), (int) path.size(), wide.data(), wlen);
-
-    if (wide.rfind(L"\\\\?\\", 0) == 0) {
-        return wide;
-    }
-
-    // GetFullPathNameW resolves relative paths and normalizes separators
-    DWORD size = GetFullPathNameW(wide.c_str(), 0, nullptr, nullptr);
-    if (size == 0) {
-        return wide;
-    }
-    std::wstring full(size, L'\0');
-    DWORD written = GetFullPathNameW(wide.c_str(), size, full.data(), nullptr);
-    if (written == 0 || written >= size) {
-        return wide;
-    }
-    full.resize(written);
-
-    if (full.size() <= 255) {
-        return wide; // short enough: open the path exactly as given
-    }
-    if (full.rfind(L"\\\\", 0) == 0) {
-        return L"\\\\?\\UNC\\" + full.substr(2); // \\server\share -> \\?\UNC\server\share
-    }
-    return L"\\\\?\\" + full;
-}
-#endif
-
-common_file_lock::status common_file_lock::try_lock() {
-    if (lock_path.empty()) {
-        return status::acquired; // no lock path, nothing to lock
-    }
-#if defined(_WIN32)
-    if (fd < 0) {
-        // opened with a wide path: extended paths cannot go through the narrow CRT
-        std::wstring lock_path_w = as_extended_path(lock_path);
-        fd = _wopen(lock_path_w.c_str(), _O_RDWR | _O_CREAT | _O_BINARY | _O_NOINHERIT, _S_IREAD | _S_IWRITE);
-        if (fd < 0) {
-            LOG_ERR("%s: cannot open lock file: %s\n", __func__, lock_path.c_str());
-            return status::error;
-        }
-    }
-    // lock the first byte, the file itself stays empty
-    OVERLAPPED ov = {};
-    locked = LockFileEx((HANDLE) _get_osfhandle(fd),
-                        LOCKFILE_EXCLUSIVE_LOCK | LOCKFILE_FAIL_IMMEDIATELY, 0, 1, 0, &ov);
-    if (locked) {
-        return status::acquired;
-    }
-    if (GetLastError() == ERROR_LOCK_VIOLATION) {
-        return status::busy;
-    }
-    LOG_ERR("%s: file locking failed (%d)\n", __func__, (int) GetLastError());
-    return status::error;
-#else
-    if (fd < 0) {
-        int open_flags = O_RDWR | O_CREAT | O_CLOEXEC;
-#ifdef O_NOFOLLOW
-        open_flags |= O_NOFOLLOW;
-#endif
-        // group-writable, so users sharing a cache can wait on the lock
-        fd = open(lock_path.c_str(), open_flags, 0664);
-        if (fd < 0) {
-            LOG_ERR("%s: cannot open lock file: %s\n", __func__, lock_path.c_str());
-            return status::error;
-        }
-        // open() filters the mode through the umask
-        (void) fchmod(fd, 0664);
-    }
-
-    if (flock(fd, LOCK_EX | LOCK_NB) == 0) {
-        locked = true;
-        return status::acquired;
-    }
-    // EACCES is reported as contention on some filesystems
-    if (errno == EWOULDBLOCK || errno == EAGAIN || errno == EACCES) {
-        return status::busy;
-    }
-    // flock is not implemented (some NFS/FUSE mounts): report it instead of racing
-    LOG_ERR("%s: file locking failed (%s)\n", __func__, strerror(errno));
-    return status::error;
-#endif
-}
-
-bool common_file_lock::acquire(const std::function<bool()> & keep_waiting) {
-    // the lock file is opened by try_lock, its parent must exist by then
-    if (!lock_path.empty()) {
-        std::error_code ec;
-        auto lock_dir = std::filesystem::path(lock_path).parent_path();
-        if (!lock_dir.empty()) {
-            std::filesystem::create_directories(lock_dir, ec);
-            if (ec) {
-                LOG_ERR("%s: unable to create lock directory '%s': %s\n", __func__, lock_dir.string().c_str(), ec.message().c_str());
-                return false;
-            }
-        }
-    }
-
-    while (true) {
-        switch (try_lock()) {
-            case status::acquired:
-                return true;
-            case status::busy:
-                if (keep_waiting && !keep_waiting()) {
-                    return false;
-                }
-                std::this_thread::sleep_for(std::chrono::milliseconds(100));
-                continue;
-            case status::error:
-                return false;
-        }
-    }
-}
-
-// the lock file is never deleted: another process could then create a fresh
-// file at the same path and lock that
-void common_file_lock::close() {
-#if defined(_WIN32)
-    if (fd >= 0) {
-        if (locked) {
-            OVERLAPPED ov = {};
-            UnlockFileEx((HANDLE) _get_osfhandle(fd), 0, 1, 0, &ov);
-        }
-        _close(fd);
-    }
-#else
-    if (fd >= 0) {
-        if (locked) {
-            flock(fd, LOCK_UN);
-        }
-        ::close(fd);
-    }
-#endif
-    fd     = -1;
-    locked = false;
-}
-
-common_file_lock::~common_file_lock() {
-    close();
-}
@@ -1,36 +0,0 @@
-#pragma once
-
-#include <functional>
-#include <string>
-
-// Cross-process exclusive lock (flock on POSIX, LockFileEx on Windows).
-// Lazy: acquire() opens the lock file, so a read-only cache works if the lock is never acquired.
-class common_file_lock {
-public:
-    common_file_lock(const std::string & lock_path);
-    ~common_file_lock();
-
-    // take the lock, waiting while another process holds it;
-    // keep_waiting is polled before each retry, false cancels the wait.
-    // Also returns false when locking is unavailable.
-    bool acquire(const std::function<bool()> & keep_waiting = {});
-
-    void close();
-
-    common_file_lock(const common_file_lock &) = delete;
-    common_file_lock & operator=(const common_file_lock &) = delete;
-
-private:
-    enum class status {
-        acquired,
-        busy,  // another process holds the lock
-        error, // file locking is unavailable: no exclusion is possible
-    };
-
-    status try_lock();
-
-    std::string lock_path;
-
-    int  fd     = -1;
-    bool locked = false;
-};
@@ -1,8 +1,6 @@
 #include "fit.h"

-#include "common.h"
 #include "json.h"
-#include "llama-cpp.h"
 #include "log.h"

 #include "../src/llama-ext.h"
@@ -29,32 +27,6 @@ class common_params_fit_exception : public std::runtime_error {
    using std::runtime_error::runtime_error;
 };

-// RAII: send llama logs below min_level to the debug log
-struct common_fit_log_scope {
-    common_fit_log_scope(ggml_log_level min_level) : min_level(min_level) {
-        llama_log_get(&original_logger.callback, &original_logger.user_data);
-        llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
-            const common_fit_log_scope * self = (const common_fit_log_scope *) user_data;
-            const ggml_log_level level_eff = level >= self->min_level ? level : GGML_LOG_LEVEL_DEBUG;
-            self->original_logger.callback(level_eff, text, self->original_logger.user_data);
-        }, this);
-    }
-
-    ~common_fit_log_scope() {
-        llama_log_set(original_logger.callback, original_logger.user_data);
-    }
-
-    common_fit_log_scope(const common_fit_log_scope &) = delete;
-    common_fit_log_scope & operator=(const common_fit_log_scope &) = delete;
-
-private:
-    struct {
-        ggml_log_callback callback;
-        void * user_data;
-    } original_logger;
-    ggml_log_level min_level;
-};
-
 static std::vector<llama_device_memory_data> common_get_device_memory_data_impl(
        const char * path_model,
        const llama_model_params * mparams,
@@ -64,7 +36,22 @@ static std::vector<llama_device_memory_data> common_get_device_memory_data_impl(
        uint32_t & hp_n_ctx_train,
        uint32_t & hp_n_expert,
        ggml_log_level log_level) {
-    common_fit_log_scope log_scope(log_level);
+    struct user_data_t {
+        struct {
+            ggml_log_callback callback;
+            void * user_data;
+        } original_logger;
+        ggml_log_level min_level; // prints below this log level go to debug log
+    };
+    user_data_t ud;
+    llama_log_get(&ud.original_logger.callback, &ud.original_logger.user_data);
+    ud.min_level = log_level;
+
+    llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
+        const user_data_t * ud = (const user_data_t *) user_data;
+        const ggml_log_level level_eff = level >= ud->min_level ? level : GGML_LOG_LEVEL_DEBUG;
+        ud->original_logger.callback(level_eff, text, ud->original_logger.user_data);
+    }, &ud);

    llama_model_params mparams_copy = *mparams;
    mparams_copy.no_alloc  = true;
@@ -72,12 +59,14 @@ static std::vector<llama_device_memory_data> common_get_device_memory_data_impl(

    llama_model * model = llama_model_load_from_file(path_model, mparams_copy);
    if (model == nullptr) {
+        llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
        throw std::runtime_error("failed to load model");
    }

    llama_context * ctx = llama_init_from_model(model, *cparams);
    if (ctx == nullptr) {
        llama_model_free(model);
+        llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
        throw std::runtime_error("failed to create llama_context from model");
    }

@@ -159,81 +148,11 @@ static std::vector<llama_device_memory_data> common_get_device_memory_data_impl(

    llama_free(ctx);
    llama_model_free(model);
+    llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);

    return ret;
 }

-std::vector<common_memory_breakdown_row> common_fit_measure(
-        const char * path_model,
-        const llama_model_params * mparams,
-        const llama_context_params * cparams,
-        uint32_t * n_ctx_train,
-        uint32_t * n_ctx,
-        ggml_log_level log_level) {
-    common_fit_log_scope log_scope(log_level);
-
-    llama_model_params mparams_copy = *mparams;
-    mparams_copy.no_alloc  = true;
-    mparams_copy.load_mode = LLAMA_LOAD_MODE_NONE;
-
-    llama_model_ptr model(llama_model_load_from_file(path_model, mparams_copy));
-    if (!model) {
-        throw std::runtime_error("failed to load model");
-    }
-
-    llama_context_ptr ctx(llama_init_from_model(model.get(), *cparams));
-    if (!ctx) {
-        throw std::runtime_error("failed to create llama_context from model");
-    }
-
-    if (n_ctx_train) {
-        *n_ctx_train = llama_model_n_ctx_train(model.get());
-    }
-    if (n_ctx) {
-        *n_ctx = llama_n_ctx(ctx.get());
-    }
-
-    return common_memory_breakdown_rows(ctx.get());
-}
-
-common_memory_reports common_measure_memory(common_params & params) {
-    const ggml_log_level log_level = params.verbosity >= LOG_LEVEL_DEBUG ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR;
-    const char * path_model = params.model.path.c_str();
-
-    common_memory_reports res;
-
-    auto mparams = common_model_params_to_llama(params);
-    auto cparams = common_context_params_to_llama(params);
-
-    // max: what --fit picks now, no-op with -fit off
-    {
-        auto mparams_fit = mparams;
-        auto cparams_fit = cparams;
-        common_fit_apply(params, mparams_fit, cparams_fit);
-        res.info_max = common_fit_measure(path_model, &mparams_fit, &cparams_fit, &res.n_ctx_train, &res.n_ctx_max, log_level);
-    }
-
-    // min: the smallest context --fit may pick (--fit only changes n_ctx when it is 0)
-    uint32_t n_ctx_min = cparams.n_ctx;
-    if (n_ctx_min == 0 && params.fit_params) {
-        const uint32_t n_streams = cparams.kv_unified ? 1 : std::max<uint32_t>(1, cparams.n_seq_max);
-        // -c 0 sets fit_params_min_ctx to UINT32_MAX, cap at n_ctx_train like --fit
-        n_ctx_min = (uint32_t) std::min<uint64_t>(
-            uint64_t(params.fit_params_min_ctx) * n_streams,
-            uint64_t(res.n_ctx_train) * n_streams);
-    }
-    if (n_ctx_min == 0 || n_ctx_min >= res.n_ctx_max) {
-        // nothing to reduce, or -fit off with the trained context
-        res.n_ctx_min = res.n_ctx_max;
-        res.info_min  = res.info_max;
-    } else {
-        cparams.n_ctx = n_ctx_min;
-        res.info_min  = common_fit_measure(path_model, &mparams, &cparams, nullptr, &res.n_ctx_min, log_level);
-    }
-
-    return res;
-}
-
 common_device_memory_data_vec common_get_device_memory_data(
        const char * path_model,
        const llama_model_params * mparams,
@@ -984,7 +903,8 @@ enum common_params_fit_status common_fit_params(
    return status;
 }

-std::vector<common_memory_breakdown_row> common_memory_breakdown_rows(const struct llama_context * ctx) {
+void common_memory_breakdown_print(const struct llama_context * ctx) {
+    //const auto & devices = ctx->get_model().devices;
    const auto * model = llama_get_model(ctx);

    std::vector<ggml_backend_dev_t> devices;
@@ -994,6 +914,18 @@ std::vector<common_memory_breakdown_row> common_memory_breakdown_rows(const stru

    llama_memory_breakdown memory_breakdown = llama_get_memory_breakdown(ctx);

+    std::vector<std::array<std::string, 9>> table_data;
+    table_data.reserve(devices.size());
+
+    // same data as the table below, for --log-jsonl consumers
+    common_json rows = common_json::array();
+    const std::string template_header = "%s: | %s | %s   %s    %s   %s   %s   %s    %s |\n";
+    const std::string template_gpu    = "%s: | %s | %s = %s + (%s = %s + %s + %s) + %s |\n";
+    const std::string template_other  = "%s: | %s | %s   %s    %s = %s + %s + %s    %s |\n";
+
+    table_data.push_back({template_header, "memory breakdown [MiB]", "total", "free", "self", "model", "context", "compute", "unaccounted"});
+
+    constexpr size_t MiB = 1024 * 1024;
    const std::vector<std::string> desc_prefixes_strip = {"NVIDIA ", "GeForce ", "Tesla ", "AMD ", "Radeon ", "Instinct "};

    // track seen buffer types to avoid double counting:
@@ -1032,130 +964,104 @@ std::vector<common_memory_breakdown_row> common_memory_breakdown_rows(const stru
        }
    }

-    std::vector<common_memory_breakdown_row> rows;
-    rows.reserve(devices.size() + 1 + memory_breakdown.size());
-
+    // print memory breakdown for each device:
    for (size_t i = 0; i < devices.size(); i++) {
        ggml_backend_dev_t dev = devices[i];
+        llama_memory_breakdown_data mb = mb_dev[i];

-        common_memory_breakdown_row row;
-        row.kind        = common_memory_breakdown_row::DEVICE;
-        row.name        = ggml_backend_dev_name(dev);
-        row.description = ggml_backend_dev_description(dev);
+        const std::string name = ggml_backend_dev_name(dev);
+        std::string desc = ggml_backend_dev_description(dev);
        for (const std::string & prefix : desc_prefixes_strip) {
-            if (row.description.length() >= prefix.length() && row.description.substr(0, prefix.length()) == prefix) {
-                row.description = row.description.substr(prefix.length());
+            if (desc.length() >= prefix.length() && desc.substr(0, prefix.length()) == prefix) {
+                desc = desc.substr(prefix.length());
            }
        }

        size_t free, total;
        ggml_backend_dev_memory(dev, &free, &total);
-        row.total   = total;
-        row.free    = free;
-        row.model   = mb_dev[i].model;
-        row.context = mb_dev[i].context;
-        row.compute = mb_dev[i].compute;
-        rows.push_back(std::move(row));
+
+        const size_t self = mb.model + mb.context + mb.compute;
+        const int64_t unaccounted = static_cast<int64_t>(total) - static_cast<int64_t>(free) - static_cast<int64_t>(self);
+
+        table_data.push_back({
+            template_gpu,
+            "  - " + name + " (" + desc + ")",
+            std::to_string(total / MiB),
+            std::to_string(free / MiB),
+            std::to_string(self / MiB),
+            std::to_string(mb.model / MiB),
+            std::to_string(mb.context / MiB),
+            std::to_string(mb.compute / MiB),
+            std::to_string(unaccounted / static_cast<int64_t>(MiB))});
+
+        rows.push_back({
+            {"kind",        "device"},
+            {"name",        name},
+            {"description", desc},
+            {"total",       total / MiB},
+            {"free",        free / MiB},
+            {"self",        self / MiB},
+            {"model",       mb.model / MiB},
+            {"context",     mb.context / MiB},
+            {"compute",     mb.compute / MiB},
+            {"unaccounted", unaccounted / static_cast<int64_t>(MiB)},
+        });
    }

+    // print memory breakdown for host:
    {
-        common_memory_breakdown_row row;
-        row.kind    = common_memory_breakdown_row::HOST;
-        row.name    = "Host";
-        row.model   = mb_host.model;
-        row.context = mb_host.context;
-        row.compute = mb_host.compute;
-        rows.push_back(std::move(row));
+        const size_t self = mb_host.model + mb_host.context + mb_host.compute;
+        table_data.push_back({
+            template_other,
+            "  - Host",
+            "", // total
+            "", // free
+            std::to_string(self / MiB),
+            std::to_string(mb_host.model / MiB),
+            std::to_string(mb_host.context / MiB),
+            std::to_string(mb_host.compute / MiB),
+            ""}); // unaccounted
+
+        rows.push_back({
+            {"kind",    "host"},
+            {"name",    "Host"},
+            {"self",    self / MiB},
+            {"model",   mb_host.model / MiB},
+            {"context", mb_host.context / MiB},
+            {"compute", mb_host.compute / MiB},
+        });
    }

-    // all remaining buffer types:
+    // print memory breakdown for all remaining buffer types:
    for (const auto & buft_mb : memory_breakdown) {
        ggml_backend_buffer_type_t          buft = buft_mb.first;
        const llama_memory_breakdown_data & mb   = buft_mb.second;
        if (seen_buffer_types.count(buft) == 1) {
            continue;
        }
-        common_memory_breakdown_row row;
-        row.kind    = common_memory_breakdown_row::BUFFER_TYPE;
-        row.name    = ggml_backend_buft_name(buft);
-        row.model   = mb.model;
-        row.context = mb.context;
-        row.compute = mb.compute;
-        rows.push_back(std::move(row));
-
-        seen_buffer_types.insert(buft);
-    }
-
-    return rows;
-}
-
-void common_memory_breakdown_print(const struct llama_context * ctx) {
-    const std::vector<common_memory_breakdown_row> rows = common_memory_breakdown_rows(ctx);
-
-    std::vector<std::array<std::string, 9>> table_data;
-    table_data.reserve(rows.size() + 1);
-
-    // same data as the table below, for --log-jsonl consumers
-    common_json json_rows = common_json::array();
-    const std::string template_header = "%s: | %s | %s   %s    %s   %s   %s   %s    %s |\n";
-    const std::string template_gpu    = "%s: | %s | %s = %s + (%s = %s + %s + %s) + %s |\n";
-    const std::string template_other  = "%s: | %s | %s   %s    %s = %s + %s + %s    %s |\n";
-
-    table_data.push_back({template_header, "memory breakdown [MiB]", "total", "free", "self", "model", "context", "compute", "unaccounted"});
-
-    constexpr size_t MiB = 1024 * 1024;
-
-    for (const auto & row : rows) {
-        const size_t self = row.model + row.context + row.compute;
-
-        if (row.kind == common_memory_breakdown_row::DEVICE) {
-            const int64_t unaccounted = row.total - row.free - static_cast<int64_t>(self);
-
-            table_data.push_back({
-                template_gpu,
-                "  - " + row.name + " (" + row.description + ")",
-                std::to_string(row.total / MiB),
-                std::to_string(row.free / MiB),
-                std::to_string(self / MiB),
-                std::to_string(row.model / MiB),
-                std::to_string(row.context / MiB),
-                std::to_string(row.compute / MiB),
-                std::to_string(unaccounted / static_cast<int64_t>(MiB))});
-
-            json_rows.push_back({
-                {"kind",        "device"},
-                {"name",        row.name},
-                {"description", row.description},
-                {"total",       row.total / MiB},
-                {"free",        row.free / MiB},
-                {"self",        self / MiB},
-                {"model",       row.model / MiB},
-                {"context",     row.context / MiB},
-                {"compute",     row.compute / MiB},
-                {"unaccounted", unaccounted / static_cast<int64_t>(MiB)},
-            });
-            continue;
-        }
-
+        const std::string name = ggml_backend_buft_name(buft);
+        const size_t self = mb.model + mb.context + mb.compute;
        table_data.push_back({
            template_other,
-            "  - " + row.name,
+            "  - " + name,
            "", // total
            "", // free
            std::to_string(self / MiB),
-            std::to_string(row.model / MiB),
-            std::to_string(row.context / MiB),
-            std::to_string(row.compute / MiB),
+            std::to_string(mb.model / MiB),
+            std::to_string(mb.context / MiB),
+            std::to_string(mb.compute / MiB),
            ""}); // unaccounted

-        json_rows.push_back({
-            {"kind",    row.kind == common_memory_breakdown_row::HOST ? "host" : "buffer_type"},
-            {"name",    row.name},
+        rows.push_back({
+            {"kind",    "buffer_type"},
+            {"name",    name},
            {"self",    self / MiB},
-            {"model",   row.model / MiB},
-            {"context", row.context / MiB},
-            {"compute", row.compute / MiB},
+            {"model",   mb.model / MiB},
+            {"context", mb.context / MiB},
+            {"compute", mb.compute / MiB},
        });
+
+        seen_buffer_types.insert(buft);
    }

    for (size_t j = 1; j < table_data[0].size(); j++) {
@@ -1175,7 +1081,7 @@ void common_memory_breakdown_print(const struct llama_context * ctx) {

    LOG_JSON("fit_memory_breakdown", common_json({
        {"unit", "MiB"},
-        {"rows", json_rows},
+        {"rows", rows},
    }));
 }

@@ -3,7 +3,6 @@
 #include "ggml.h"
 #include "llama.h"

-#include <string>
 #include <vector>

 enum common_params_fit_status {
@@ -46,47 +45,6 @@ void common_fit_print(

 void common_memory_breakdown_print(const llama_context * ctx);

-// one row of the memory breakdown: a device, the host or another buffer type
-struct common_memory_breakdown_row {
-    enum kind_t { DEVICE, HOST, BUFFER_TYPE };
-
-    kind_t      kind = DEVICE;
-    std::string name;        // device or buffer type name, "Host" for the host row
-    std::string description; // device description, empty otherwise
-    int64_t     total = 0;   // device memory in bytes, 0 unless kind == DEVICE
-    int64_t     free  = 0;   // free device memory in bytes, 0 unless kind == DEVICE
-    size_t      model   = 0; // bytes held by this context, by kind of buffer
-    size_t      context = 0;
-    size_t      compute = 0;
-};
-
-// memory a model needs at the min and max context --fit may pick
-struct common_memory_reports {
-    uint32_t n_ctx_train = 0;
-    uint32_t n_ctx_max   = 0; // the largest context that fits the free device memory right now
-    uint32_t n_ctx_min   = 0; // --fit-ctx (the smallest context --fit may pick), with the placement set in params
-    std::vector<common_memory_breakdown_row> info_max; // breakdown at n_ctx_max
-    std::vector<common_memory_breakdown_row> info_min; // breakdown at n_ctx_min
-};
-
-// rows of common_memory_breakdown_print: model devices first, then host
-std::vector<common_memory_breakdown_row> common_memory_breakdown_rows(const llama_context * ctx);
-
-// load model + context with no_alloc and return the memory breakdown, params are used as-is
-// n_ctx_train and n_ctx are optional outputs; throws on failure
-std::vector<common_memory_breakdown_row> common_fit_measure(
-                         const char * path_model,
-           const llama_model_params * mparams,
-         const llama_context_params * cparams,
-                           uint32_t * n_ctx_train,
-                           uint32_t * n_ctx,
-                     ggml_log_level   log_level);
-
-struct common_params;
-
-// measure min and max without allocating, params may be modified as with --fit; throws on failure
-common_memory_reports common_measure_memory(common_params & params);
-
 struct common_device_memory_data {
    int64_t total;
    int64_t free;
@@ -2,7 +2,6 @@

 #include "build-info.h"
 #include "common.h"
-#include "file-lock.h"
 #include "log.h"
 #include "http.h"
 #include "json.h"
@@ -63,7 +62,7 @@ static fs::path get_cache_directory() {
 }

 std::string get_cache_path() {
-    return get_cache_directory().string();
+    return fs_path_to_utf8(get_cache_directory());
 }

 static std::string folder_name_to_repo(const std::string & folder) {
@@ -162,14 +161,6 @@ static bool is_valid_oid(const std::string & oid) {
    return is_hex_string(oid, 40) || is_hex_string(oid, 64);
 }

-std::string get_lock_path(const std::string & repo_id, const std::string & oid) {
-    if (!is_valid_repo_id(repo_id) || !is_valid_oid(oid)) {
-        return {};
-    }
-    fs::path lock_path = fs::path(get_cache_path()) / ".locks" / repo_to_folder_name(repo_id) / (oid + ".lock");
-    return lock_path.string();
-}
-
 static bool is_valid_subpath(const fs::path & path, const fs::path & subpath) {
    if (subpath.is_absolute()) {
        return false; // never do a / b with b absolute
@@ -340,11 +331,11 @@ hf_files get_repo_files(const std::string & repo_id,
            file.url = endpoint + repo_id + "/resolve/" + commit + "/" + file.path;

            fs::path final_path = commit_path / subpath;
-            file.final_path = final_path.string();
+            file.final_path = fs_path_to_utf8(final_path);

            if (!file.oid.empty() && !fs::exists(final_path)) {
                fs::path local_path = blobs_path / file.oid;
-                file.local_path = local_path.string();
+                file.local_path = fs_path_to_utf8(local_path);
            } else {
                file.local_path = file.final_path;
            }
@@ -411,7 +402,7 @@ hf_files get_cached_files(const std::string & repo_id) {
        if (!fs::exists(snapshots_path)) {
            continue;
        }
-        std::string _repo_id = folder_name_to_repo(repo.path().filename().string());
+        std::string _repo_id = folder_name_to_repo(fs_path_to_utf8(repo.path().filename()));

        if (!is_valid_repo_id(_repo_id)) {
            continue;
@@ -436,7 +427,7 @@ hf_files get_cached_files(const std::string & repo_id) {
                file.repo_id = _repo_id;
                const auto generic_path = path.generic_u8string();
                file.path = std::string(generic_path.begin(), generic_path.end());
-                file.local_path = entry.path().string();
+                file.local_path = fs_path_to_utf8(entry.path());
                file.final_path = file.local_path;
                files.push_back(std::move(file));
            }
@@ -446,7 +437,7 @@ hf_files get_cached_files(const std::string & repo_id) {
    return files;
 }

-std::string finalize_file(const hf_file & file, const std::function<bool()> & keep_waiting) {
+std::string finalize_file(const hf_file & file) {
    static std::atomic<bool> symlinks_disabled{false};

    std::error_code ec;
@@ -461,24 +452,6 @@ std::string finalize_file(const hf_file & file, const std::function<bool()> & ke
        return file.final_path;
    }

-    // the lock file is shared with huggingface_hub: another process may be
-    // finalizing the same blob
-    common_file_lock lock(get_lock_path(file.repo_id, file.oid));
-    if (!lock.acquire(keep_waiting)) {
-        // without the lock, another process could move the blob while we check it
-        throw std::runtime_error("failed to lock cache for finalization: " + file.final_path);
-    }
-
-    // acquire() only polls keep_waiting while waiting for a busy lock
-    if (keep_waiting && !keep_waiting()) {
-        throw std::runtime_error("download cancelled before finalization: " + file.final_path);
-    }
-
-    // re-check under the lock: the winner may have created the link already
-    if (fs::exists(final_path, ec)) {
-        return file.final_path;
-    }
-
    fs::create_directories(final_path.parent_path(), ec);

    if (!symlinks_disabled) {
@@ -516,7 +489,7 @@ bool remove_cached_repo(const std::string & repo_id) {
    std::error_code ec;
    auto removed = fs::remove_all(repo_path, ec);
    if (ec) {
-        LOG_ERR("%s: failed to remove repo cache %s: %s\n", __func__, repo_path.string().c_str(), ec.message().c_str());
+        LOG_ERR("%s: failed to remove repo cache %s: %s\n", __func__, fs_path_to_utf8(repo_path).c_str(), ec.message().c_str());
        return false;
    }
    return removed > 0;
@@ -1,6 +1,5 @@
 #pragma once

-#include <functional>
 #include <string>
 #include <vector>

@@ -27,14 +26,8 @@ hf_files get_repo_files(

 hf_files get_cached_files(const std::string & repo_id = {});

-// Lock file path for a blob, shared with the hf CLI: <cache>/.locks/<repo folder>/<oid>.lock
-// Pure path computation: creates nothing. Returns an empty string on invalid repo or oid.
-std::string get_lock_path(const std::string & repo_id, const std::string & oid);
-
-// Create snapshot path (link or move/copy) and return it.
-// Throws when the lock cannot be taken or keep_waiting returns false.
-// A completed snapshot (final_path present) is returned without the lock, so a read-only cache works.
-std::string finalize_file(const hf_file & file, const std::function<bool()> & keep_waiting = {});
+// Create snapshot path (link or move/copy) and return it
+std::string finalize_file(const hf_file & file);

 // Remove the entire cached directory for a repo, returns true if removed
 bool remove_cached_repo(const std::string & repo_id);
@@ -6,9 +6,7 @@
 // we can remove it in the future when we figure out a better way to define inputs using jinja::value
 #include "json.h"

-#include <algorithm>
 #include <functional>
-#include <optional>
 #include <sstream>

 #define FILENAME "jinja-caps"
@@ -186,10 +184,6 @@ caps caps_get(jinja::program & prog) {
                    {
                        {"role", "user"},
                        {"content", json::array({
-                            {
-                                {"type", "text"},
-                                {"text", content_marker}
-                            }
                        })}
                    }
                });
@@ -590,58 +584,6 @@ caps caps_get(jinja::program & prog) {
        }
    );

-    if (result.supports_reasoning_effort) {
-        JJ_DEBUG("%s\n", ">>> Running capability check: reasoning effort levels");
-
-        static const std::vector<std::string> candidates = { "minimal", "low", "medium", "high", "xhigh", "max" };
-
-        // renders a single user turn with the given effort, nothing when the template raises
-        const auto render = [&](const std::string & effort) {
-            std::optional<std::string> rendered;
-            caps_try_execute(
-                prog,
-                [&]() {
-                    // messages
-                    return json::array({
-                        {
-                            {"role", "user"},
-                            {"content", "User message"}
-                        },
-                    });
-                },
-                [&](context & ctx) {
-                    ctx.set_val("enable_thinking", mk_val<value_bool>(true));
-                    caps_apply_reasoning_effort(ctx, effort);
-                },
-                nullptr, // tools_fn
-                [&](context &, bool success, value &, value &, const std::string & output) {
-                    if (success) {
-                        rendered = output;
-                    }
-                }
-            );
-            return rendered;
-        };
-
-        // an unknown level shows the render of the template fallback, a level that renders the same is not honored
-        // levels sharing a render are aliases and the highest one is kept
-        const auto unknown = render("REASONING_EFFORT_UNKNOWN");
-        std::vector<std::string> renders;
-        for (const auto & level : candidates) {
-            const auto rendered = render(level);
-            if (!rendered || rendered == unknown) {
-                continue;
-            }
-            const auto it = std::find(renders.begin(), renders.end(), *rendered);
-            if (it != renders.end()) {
-                result.reasoning_effort_levels[it - renders.begin()] = level;
-            } else {
-                renders.push_back(*rendered);
-                result.reasoning_effort_levels.push_back(level);
-            }
-        }
-    }
-
    JJ_DEBUG("%s\n", result.to_string().c_str());

    return result;
@@ -4,7 +4,6 @@

 #include <string>
 #include <map>
-#include <vector>

 namespace jinja {

@@ -20,9 +19,6 @@ struct caps {
    // supports reasoning effort levels
    bool supports_reasoning_effort = false;

-    // reasoning effort levels the template renders distinctly, from lowest to highest
-    std::vector<std::string> reasoning_effort_levels;
-
    // one of the 2 content capabilities must be true
    bool supports_string_content = true;
    bool supports_typed_content = false;
@@ -1,9 +1,7 @@
 #include "lexer.h"
 #include "runtime.h"
-#include "unicode.h"

 #include <cctype>
-#include <charconv>
 #include <functional>
 #include <map>
 #include <string>
@@ -77,24 +75,6 @@ lexer_result lexer::tokenize(const std::string & source) {
                }
                // add escaped char
                char escaped_char = src[pos++];
-                if (escaped_char == 'x') {
-                    if (pos + 1 >= src.size()) {
-                        throw lexer_exception("incomplete hexadecimal escape", source, pos);
-                    }
-
-                    unsigned int value = 0;
-                    const char * first = &src[pos];
-                    const char * last  = first + 2;
-
-                    auto [ptr, ec] = std::from_chars(first, last, value, 16);
-                    if (ec != std::errc{} || ptr != last) {
-                        throw lexer_exception("invalid hexadecimal escape", source, pos);
-                    }
-
-                    str += common_unicode_cpt_to_utf8(value);
-                    pos += 2;
-                    continue;
-                }
                if (escape_chars.find(escaped_char) == escape_chars.end()) {
                    throw lexer_exception(std::string("unknown escape character \\") + escaped_char, source, pos);
                }
@@ -21,17 +21,6 @@ static bool is_type(const statement_ptr & ptr) {
 class parser {
    const std::vector<token> & tokens;
    size_t current = 0;
-    int depth = 0;
-
-    struct depth_guard {
-        parser & p;
-        depth_guard(parser & p) : p(p) {
-            if (++p.depth > JINJA_RECURSIVE_LIMIT) {
-                throw parser_exception("Parser Error: template is nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep", p.source, p.peek().pos);
-            }
-        }
-        ~depth_guard() { p.depth--; }
-    };

    std::string source; // for error reporting

@@ -105,7 +94,6 @@ private:
    }

    statement_ptr parse_any() {
-        depth_guard guard(*this);
        size_t start_pos = current;
        switch (peek().t) {
            case token::comment:
@@ -266,7 +254,6 @@ private:
            size_t pos0 = current;
            ++current; // consume {%
            ++current; // consume 'elif'
-            depth_guard guard(*this);
            alternate.push_back(parse_if_statement(pos0)); // nested If
        } else if (is_statement({"else"})) {
            ++current; // consume {%
@@ -343,7 +330,6 @@ private:
    }

    statement_ptr parse_if_expression() {
-        depth_guard guard(*this);
        auto a = parse_logical_or_expression();
        if (is_identifier("if")) {
            // Ternary expression
@@ -387,7 +373,6 @@ private:
    statement_ptr parse_logical_negation_expression() {
        // Try parse unary operators
        if (is_identifier("not")) {
-            depth_guard guard(*this);
            size_t start_pos = current;
            auto op = next();
            return mk_stmt<unary_expression>(start_pos, op, parse_logical_negation_expression());
@@ -473,7 +458,6 @@ private:

    statement_ptr parse_unary_expression() {
        if (is(token::unary_operator)) {
-            depth_guard guard(*this);
            size_t start_pos = current;
            auto op = next();
            return mk_stmt<unary_expression>(start_pos, op, parse_unary_expression());
@@ -493,11 +477,9 @@ private:
        size_t start_pos = current;
        auto expr = mk_stmt<call_expression>(start_pos, std::move(callee), parse_args());
        auto member = parse_member_expression(std::move(expr)); // foo.x().y
-        if (!is(token::open_paren)) {
-            return member;
-        }
-        depth_guard guard(*this);
-        return parse_call_expression(std::move(member)); // foo.x()()
+        return is(token::open_paren)
+            ? parse_call_expression(std::move(member)) // foo.x()()
+            : std::move(member);
    }

    statements parse_args() {
@@ -3,7 +3,6 @@
 #include "value.h"
 #include "utils.h"

-#include <exception>
 #include <string>
 #include <vector>
 #include <memory>
@@ -53,39 +52,32 @@ static void ensure_key_type_allowed(const value & val) {

 // execute with error handling
 value statement::execute(context & ctx) const {
-    std::exception_ptr pending;
-    std::string message;
    try {
        return execute_impl(ctx);
    } catch (const continue_statement::signal & /* ex */) {
-        pending = std::current_exception();
+        throw;
    } catch (const break_statement::signal & /* ex */) {
-        pending = std::current_exception();
+        throw;
    } catch (const rethrown_exception & /* ex */) {
-        pending = std::current_exception();
+        throw;
    } catch (const not_implemented_exception & /* ex */) {
-        pending = std::current_exception();
+        throw;
    } catch (const std::exception & e) {
        const std::string & source = *ctx.src;
        if (source.empty()) {
            std::ostringstream oss;
            oss << "\nError executing " << type() << " at position " << pos << ": " << e.what();
-            message = oss.str();
+            throw rethrown_exception(oss.str());
        } else {
            std::ostringstream oss;
            oss << "\n------------\n";
            oss << "While executing " << type() << " at " << get_line_col(source, pos) << " in source:\n";
            oss << peak_source(source, pos) << "\n";
            oss << "Error: " << e.what();
-            message = oss.str();
+            // throw as another exception to avoid repeated formatting
+            throw rethrown_exception(oss.str());
        }
    }
-    // throw outside the catch: MSVC keeps the unwound frames on the stack while a catch block runs
-    if (pending) {
-        std::rethrow_exception(pending);
-    }
-    // throw as another exception to avoid repeated formatting
-    throw rethrown_exception(message);
 }

 value identifier::execute_impl(context & ctx) const {
@@ -755,11 +747,7 @@ value macro_statement::execute_impl(context & ctx) const {
    std::string name = cast_stmt<identifier>(this->name)->val;

    const func_handler func = [this, name](const func_args & args) -> value {
-        if (args.ctx.macro_depth >= JINJA_RECURSIVE_LIMIT) {
-            throw std::runtime_error("macro calls are nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
-        }
        context macro_ctx(args.ctx); // new scope for macro execution
-        macro_ctx.macro_depth = args.ctx.macro_depth + 1;

        bind_parameters(name, this->args, args, macro_ctx);

@@ -790,11 +778,7 @@ value call_statement::execute_impl(context & ctx) const {
    context caller_ctx(ctx); // new scope for caller execution

    const func_handler func = [this, caller_ctx = std::move(caller_ctx)](const func_args & args) -> value {
-        if (args.ctx.macro_depth >= JINJA_RECURSIVE_LIMIT) {
-            throw std::runtime_error("macro calls are nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
-        }
        context block_ctx(caller_ctx); // new scope for block execution
-        block_ctx.macro_depth = args.ctx.macro_depth + 1;

        bind_parameters("caller", this->caller_args, args, block_ctx);

@@ -58,8 +58,6 @@ struct context {

    bool is_get_stats = false; // whether to collect stats

-    int macro_depth = 0;
-
    visitor_fn visitor;

    // src is optional, used for error reporting
@@ -80,7 +78,6 @@ struct context {
        src(parent.src),
        current_time(parent.current_time),
        is_get_stats(parent.is_get_stats),
-        macro_depth(parent.macro_depth),
        // inherit variables (for example, when entering a new scope)
        env(mk_val<value_object>(*parent.env)) {}

@@ -713,7 +710,7 @@ struct rethrown_exception : public std::exception {

 //////////////////////

-static void gather_string_parts_recursive(const value & val, value_string & parts, int depth = 0) {
+static void gather_string_parts_recursive(const value & val, value_string & parts) {
    // TODO: probably allow print value_none as "None" string? currently this breaks some templates
    if (is_val<value_string>(val)) {
        const auto & str_val = cast_val<value_string>(val)->val_str;
@@ -721,12 +718,10 @@ static void gather_string_parts_recursive(const value & val, value_string & part
    } else if (is_val<value_int>(val) || is_val<value_float>(val) || is_val<value_bool>(val)) {
        std::string str_val = val->as_string().str();
        parts->val_str.append(str_val);
-    } else if (is_val<value_array>(val) && depth >= JINJA_RECURSIVE_LIMIT) {
-        throw std::runtime_error("value is nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
    } else if (is_val<value_array>(val)) {
        auto items = cast_val<value_array>(val)->as_array();
        for (const auto & item : items) {
-            gather_string_parts_recursive(item, parts, depth + 1);
+            gather_string_parts_recursive(item, parts);
        }
    }
 }
@@ -6,8 +6,6 @@
 #include <cstdint>
 #include <cstring>

-#define JINJA_RECURSIVE_LIMIT 128
-
 namespace jinja {

 static void string_replace_all(std::string & s, const std::string & search, const std::string & replace) {
@@ -6,7 +6,6 @@
 #include "json.h"

 #include <sstream>
-#include <stdexcept>
 #include <string>
 #include <cctype>
 #include <cfloat>
@@ -300,53 +299,74 @@ static value & get_attribute(const value & val, const value & attr, value & defa
    return default_val;
 }

-template<bool is_reject, bool lookup_attr>
-static value select_or_reject(const func_args & args) {
-    // example: array | select("odd") -> test_is_odd(item)
-    // example: array | selectattr("status", "equalto", "active") -> test_is_equalto(item.status, "active")
-    constexpr size_t test_pos = lookup_attr ? 2 : 1;
-    args.ensure_count(test_pos);
-    args.ensure_val<value_array>(args.get_pos(0));
-    value attr;
-    if constexpr (lookup_attr) {
-        attr = args.get_pos(1);
-        args.ensure_val<value_string>(attr);
-    }
-
-    auto & builtins = global_builtins();
-    auto test_it = builtins.end();
-    if (args.count() > test_pos) {
-        args.ensure_val<value_string>(args.get_pos(test_pos));
-        std::string test_name = args.get_pos(test_pos)->as_string().str();
-        test_it = builtins.find("test_is_" + test_name);
-        if (test_it == builtins.end()) {
-            throw raised_exception(args.func_name + ": unknown test '" + test_name + "'");
-        }
-    }
+template<bool is_reject>
+static value selectattr(const func_args & args) {
+    args.ensure_count(2, 4);
+    args.ensure_vals<value_array, value_string, value_string, value_string>(true, true, false, false);

    auto arr = args.get_pos(0)->as_array();
+    auto attribute = args.get_pos(1);
    auto out = mk_val<value_array>();
    value val_default = mk_val<value_undefined>();

-    for (const auto & item : arr) {
-        value val = item;
-        if constexpr (lookup_attr) {
-            val = get_attribute(item, attr, val_default);
+    if (args.count() == 2) {
+        // example: array | selectattr("active")
+        for (const auto & item : arr) {
+            value attr_val = get_attribute(item, attribute, val_default);
+            bool is_selected = attr_val->as_bool();
+            if constexpr (is_reject) is_selected = !is_selected;
+            if (is_selected) out->push_back(item);
        }
-        bool is_selected;
-        if (test_it == builtins.end()) {
-            is_selected = val->as_bool();
-        } else {
+        return out;
+
+    } else if (args.count() == 3) {
+        // example: array | selectattr("equalto", "text")
+        // translated to: test_is_equalto(item, "text")
+        std::string test_name = args.get_pos(1)->as_string().str();
+        value test_val = args.get_pos(2);
+        auto & builtins = global_builtins();
+        auto it = builtins.find("test_is_" + test_name);
+        if (it == builtins.end()) {
+            throw raised_exception("selectattr: unknown test '" + test_name + "'");
+        }
+        auto test_fn = it->second;
+        for (const auto & item : arr) {
            func_args test_args(args.ctx);
-            test_args.push_back(val);
-            for (size_t i = test_pos + 1; i < args.count(); i++) {
-                test_args.push_back(args.get_pos(i));
-            }
-            is_selected = test_it->second(test_args)->as_bool();
+            test_args.push_back(item); // current object
+            test_args.push_back(test_val); // extra argument
+            value test_result = test_fn(test_args);
+            bool is_selected = test_result->as_bool();
+            if constexpr (is_reject) is_selected = !is_selected;
+            if (is_selected) out->push_back(item);
        }
-        if constexpr (is_reject) is_selected = !is_selected;
-        if (is_selected) out->push_back(item);
+        return out;
+
+    } else if (args.count() == 4) {
+        // example: array | selectattr("status", "equalto", "active")
+        // translated to: test_is_equalto(item.status, "active")
+        std::string test_name = args.get_pos(2)->as_string().str();
+        auto extra_arg = args.get_pos(3);
+        auto & builtins = global_builtins();
+        auto it = builtins.find("test_is_" + test_name);
+        if (it == builtins.end()) {
+            throw raised_exception("selectattr: unknown test '" + test_name + "'");
+        }
+        auto test_fn = it->second;
+        for (const auto & item : arr) {
+            value attr_val = get_attribute(item, attribute, val_default);
+            func_args test_args(args.ctx);
+            test_args.push_back(attr_val); // attribute value
+            test_args.push_back(extra_arg); // extra argument
+            value test_result = test_fn(test_args);
+            bool is_selected = test_result->as_bool();
+            if constexpr (is_reject) is_selected = !is_selected;
+            if (is_selected) out->push_back(item);
+        }
+        return out;
+    } else {
+        throw raised_exception("selectattr: invalid number of arguments");
    }
+
    return out;
 }

@@ -405,12 +425,7 @@ const func_builtins & global_builtins() {
            throw raised_exception("Jinja Exception: " + msg);
        }},
        {"dict", toobject},
-        {"namespace", [](const func_args & args) -> value {
-            // unlike a dict, a namespace has no methods: ns.items is the attribute, as in jinja2
-            value out = toobject(args);
-            cast_val<value_object>(out)->has_builtins = false;
-            return out;
-        }},
+        {"namespace", toobject},
        {"strftime_now", [](const func_args & args) -> value {
            args.ensure_vals<value_string>();
            std::string format = args.get_pos(0)->as_string().str();
@@ -1075,10 +1090,10 @@ const func_builtins & value_array_t::get_builtins() const {
            auto arr = slice(val->as_array(), start, stop, step);
            return is_val<value_tuple>(val) ? mk_val<value_tuple>(std::move(arr)) : mk_val<value_array>(std::move(arr));
        }},
-        {"selectattr", select_or_reject<false, true>},
-        {"select", select_or_reject<false, false>},
-        {"rejectattr", select_or_reject<true, true>},
-        {"reject", select_or_reject<true, false>},
+        {"selectattr", selectattr<false>},
+        {"select", selectattr<false>},
+        {"rejectattr", selectattr<true>},
+        {"reject", selectattr<true>},
        {"join", [](const func_args & args) -> value {
            args.ensure_count(1, 3);
            if (!is_val<value_array>(args.get_pos(0))) {
@@ -1440,8 +1455,7 @@ const func_builtins & value_undefined_t::get_builtins() const {
 //////////////////////////////////


-// the runtime walks values with recursion, a deeper input overflows the stack
-static value from_json(const common_json & j, bool mark_input, int depth = 0) {
+static value from_json(const common_json & j, bool mark_input) {
    if (j.is_null()) {
        return mk_val<value_none>();
    } else if (j.is_boolean()) {
@@ -1456,18 +1470,16 @@ static value from_json(const common_json & j, bool mark_input, int depth = 0) {
            str->mark_input();
        }
        return str;
-    } else if ((j.is_array() || j.is_object()) && depth >= JINJA_RECURSIVE_LIMIT) {
-        throw std::invalid_argument("JSON is nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
    } else if (j.is_array()) {
        auto arr = mk_val<value_array>();
        for (const auto & item : j) {
-            arr->push_back(from_json(item, mark_input, depth + 1));
+            arr->push_back(from_json(item, mark_input));
        }
        return arr;
    } else if (j.is_object()) {
        auto obj = mk_val<value_object>();
        for (auto it = j.begin(); it != j.end(); ++it) {
-            obj->insert(it.key(), from_json(it.value(), mark_input, depth + 1));
+            obj->insert(it.key(), from_json(it.value(), mark_input));
        }
        return obj;
    } else {
@@ -1662,8 +1674,6 @@ static void value_to_json_internal(std::ostringstream & oss, const value & val,
            }
        }
        oss << "\"";
-    } else if ((is_val<value_array>(val) || is_val<value_object>(val)) && curr_lvl >= JINJA_RECURSIVE_LIMIT) {
-        throw std::runtime_error("value is nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
    } else if (is_val<value_array>(val)) {
        const auto & arr = val->as_array();
        oss << "[";
@@ -1729,20 +1739,17 @@ std::string value_to_string_repr(const value & val) {
 }

 // stats utility
-void value_t::stats_t::mark_used(value & val, bool deep, int depth) {
+void value_t::stats_t::mark_used(value & val, bool deep) {
    val->stats.used = true;
-    if (deep && (is_val<value_array>(val) || is_val<value_object>(val)) && depth >= JINJA_RECURSIVE_LIMIT) {
-        throw std::runtime_error("value is nested more than " + std::to_string(JINJA_RECURSIVE_LIMIT) + " levels deep");
-    }
    if (deep) {
        if (is_val<value_array>(val)) {
            for (auto & item : val->val_arr) {
-                mark_used(item, deep, depth + 1);
+                mark_used(item, deep);
            }
        } else if (is_val<value_object>(val)) {
            for (auto & pair : val->val_obj) {
-                mark_used(pair.first, deep, depth + 1);
-                mark_used(pair.second, deep, depth + 1);
+                mark_used(pair.first, deep);
+                mark_used(pair.second, deep);
            }
        }
    }
@@ -119,7 +119,7 @@ struct value_t {
        // ops can be builtin calls or operators: "array_access", "object_access"
        std::set<std::string> ops;
        // utility to recursively mark value and its children as used
-        static void mark_used(value & val, bool deep = false, int depth = 0);
+        static void mark_used(value & val, bool deep = false);
    } stats;

    value_t() = default;
@@ -272,12 +272,15 @@ static bool is_reserved_name(const std::string & name) {
    return RESERVED_NAMES.find(name) != RESERVED_NAMES.end();
 }

+static std::regex INVALID_RULE_CHARS_RE("[^a-zA-Z0-9-]+");
 static std::regex GRAMMAR_LITERAL_ESCAPE_RE("[\r\n\"\\\\]");
 static std::regex GRAMMAR_RANGE_LITERAL_ESCAPE_RE("[\r\n\"\\]\\-\\\\]");
 static std::unordered_map<char, std::string> GRAMMAR_LITERAL_ESCAPES = {
    {'\r', "\\r"}, {'\n', "\\n"}, {'"', "\\\""}, {'-', "\\-"}, {']', "\\]"}, {'\\', "\\\\"}
 };

+static const int MAX_PATTERN_DEPTH = 100;
+
 static std::unordered_set<char> NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?', '^', '$'};
 static std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};

@@ -309,22 +312,6 @@ static std::string format_literal(const std::string & literal) {

 std::string gbnf_format_literal(const std::string & literal) { return format_literal(literal); }

-// same as std::regex_replace with [^a-zA-Z0-9-]+ -> "-", no std::regex: libstdc++ recurses per char
-std::string gbnf_format_rule_name(const std::string & name) {
-    std::string out;
-    bool in_invalid = false;
-    for (char c : name) {
-        if ((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') || c == '-') {
-            out += c;
-            in_invalid = false;
-        } else if (!in_invalid) {
-            out += '-';
-            in_invalid = true;
-        }
-    }
-    return out;
-}
-
 static size_t gbnf_escape_length(const std::string & pattern, size_t pos) {
    if (pos + 1 >= pattern.length() || pattern[pos] != '\\') {
        return 0;
@@ -360,17 +347,6 @@ private:
    std::unordered_set<std::string> _refs_being_resolved;
    std::vector<std::string> _errors;
    std::vector<std::string> _warnings;
-    int _depth = 0;
-
-    struct depth_guard {
-        int & depth;
-        depth_guard(int & depth) : depth(depth) {
-            if (++depth > COMMON_RECURSIVE_LIMIT) {
-                throw std::invalid_argument("JSON schema is nested more than " + std::to_string(COMMON_RECURSIVE_LIMIT) + " levels deep");
-            }
-        }
-        ~depth_guard() { depth--; }
-    };

    template <typename T>
    static const T & as(const common_chat_schema & node) {
@@ -378,7 +354,7 @@ private:
    }

    std::string _add_rule(const std::string & name, const std::string & rule) {
-        std::string esc_name = gbnf_format_rule_name(name);
+        std::string esc_name = regex_replace(name, INVALID_RULE_CHARS_RE, "-");
        if (_rules.find(esc_name) == _rules.end() || _rules[esc_name] == rule) {
            _rules[esc_name] = rule;
            return esc_name;
@@ -506,7 +482,7 @@ private:
                        }
                    }
                    paren_depth++;
-                    if (paren_depth > COMMON_RECURSIVE_LIMIT) {
+                    if (paren_depth > MAX_PATTERN_DEPTH) {
                        throw unsupported_pattern("pattern nesting too deep");
                    }
                    seq.emplace_back("(" + to_rule(transform()) + ")", false);
@@ -668,11 +644,7 @@ private:
        std::string char_rule = _add_primitive("char", PRIMITIVE_RULES.at("char"));
        std::ostringstream out;
        out << "[\"] ( ";
-        // one level per char of the longest name
-        std::function<void(size_t, int)> visit = [&](size_t idx, int depth) {
-            if (depth >= COMMON_RECURSIVE_LIMIT) {
-                throw std::invalid_argument("JSON schema property name is longer than " + std::to_string(COMMON_RECURSIVE_LIMIT) + " chars");
-            }
+        std::function<void(size_t)> visit = [&](size_t idx) {
            const auto & node = trie.nodes[idx];
            std::string rejects;
            auto first = true;
@@ -687,7 +659,7 @@ private:
                out << "[" << c << "]";
                if (!trie.nodes[child].children.empty()) {
                    out << " (";
-                    visit(child, depth + 1);
+                    visit(child);
                    out << ")";
                } else {
                    out << " " << char_rule << "+";
@@ -697,7 +669,7 @@ private:
                out << " | [^\"" << rejects << "] " << char_rule << "*";
            }
        };
-        visit(0, 0);
+        visit(0);

        out << " )";
        if (trie.nodes[0].pattern < 0) {
@@ -710,7 +682,8 @@ private:
    std::string _resolve_ref(const common_chat_schema_ref & schema) {
        auto it = schema.ref.find('#');
        std::string ref_fragment = it != std::string::npos ? schema.ref.substr(it + 1) : schema.ref;
-        std::string ref_name = "ref" + gbnf_format_rule_name(ref_fragment);
+        static const std::regex nonalphanumeric_regex(R"([^a-zA-Z0-9-]+)");
+        std::string ref_name = "ref" + std::regex_replace(ref_fragment, nonalphanumeric_regex, "-");
        if (_rules.find(ref_name) == _rules.end() && _refs_being_resolved.find(schema.ref) == _refs_being_resolved.end()) {
            if (!schema.target) {
                _errors.push_back("Unresolved $ref " + schema.ref);
@@ -860,13 +833,10 @@ public:
        std::unordered_set<std::string> required;
        std::vector<std::pair<std::string, const common_chat_schema *>> properties;
        std::map<std::string, size_t> enum_values;
-        std::function<void(const common_chat_schema &, bool, int)> add_component = [&](const common_chat_schema & comp, bool is_required, int depth) {
-            if (depth >= COMMON_RECURSIVE_LIMIT) {
-                throw std::invalid_argument("JSON schema is nested more than " + std::to_string(COMMON_RECURSIVE_LIMIT) + " levels deep");
-            }
+        std::function<void(const common_chat_schema &, bool)> add_component = [&](const common_chat_schema & comp, bool is_required) {
            if (comp.kind() == common_chat_schema::KIND_REF) {
                if (const auto * target = as<common_chat_schema_ref>(comp).target) {
-                    add_component(*target, is_required, depth + 1);
+                    add_component(*target, is_required);
                }
            } else if (comp.kind() == common_chat_schema::KIND_OBJECT) {
                for (const auto & prop : as<common_chat_schema_object>(comp).properties) {
@@ -884,10 +854,10 @@ public:
        for (const auto & child : schema.children) {
            if (child->kind() == common_chat_schema::KIND_ANY_OF) {
                for (const auto & alt : as<common_chat_schema_any_of>(*child).children) {
-                    add_component(*alt, false, 0);
+                    add_component(*alt, false);
                }
            } else {
-                add_component(*child, true, 0);
+                add_component(*child, true);
            }
        }
        if (!enum_values.empty()) {
@@ -905,8 +875,6 @@ public:
    }

    std::string visit(const common_chat_schema & schema, const std::string & name) {
-        // the $ref targets are visited too, a $ref chain is not bounded by the builder
-        depth_guard guard(_depth);
        std::string rule_name = is_reserved_name(name) ? name + "-" : name.empty() ? "root" : name;
        std::string sub_name  = name + (name.empty() ? "" : "-");

@@ -19,6 +19,5 @@ struct common_grammar_options {
 };

 std::string gbnf_format_literal(const std::string & literal);
-std::string gbnf_format_rule_name(const std::string & name);

 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options = {});
@@ -9,18 +9,6 @@
 #include <utility>
 #include <vector>

-namespace {
-struct depth_guard {
-    int & depth;
-    depth_guard(int & depth) : depth(depth) {
-        if (++depth > COMMON_RECURSIVE_LIMIT) {
-            throw std::invalid_argument("JSON schema is nested more than " + std::to_string(COMMON_RECURSIVE_LIMIT) + " levels deep");
-        }
-    }
-    ~depth_guard() { depth--; }
-};
-} // namespace
-
 class common_chat_schema_builder {
    const common_json &      root_;
    common_chat_schema_document & doc_;
@@ -236,14 +224,7 @@ class common_chat_schema_builder {
        return node;
    }

-    int depth_ = 0;
-
    common_chat_schema_ptr build_node(const common_json & schema, const std::string & path) {
-        depth_guard guard(depth_);
-        return build_node_impl(schema, path);
-    }
-
-    common_chat_schema_ptr build_node_impl(const common_json & schema, const std::string & path) {
        if (!schema.is_object()) {
            fail(path, "schema must be an object");
        }
@@ -384,9 +365,7 @@ static common_chat_schema::value_type json_type(const common_json & value) {
    return common_chat_schema::TYPE_OBJECT;
 }

-// depth counts the $ref, anyOf and allOf hops, a $ref chain is not bounded by the builder
-static common_chat_schema::type_set value_types_impl(const common_chat_schema & s, std::unordered_set<const common_chat_schema *> & visited, int & depth) {
-    depth_guard guard(depth);
+static common_chat_schema::type_set value_types_impl(const common_chat_schema & s, std::unordered_set<const common_chat_schema *> & visited) {
    switch (s.kind()) {
        case common_chat_schema::KIND_ANY:
            return common_chat_schema::type_set::all();
@@ -420,21 +399,21 @@ static common_chat_schema::type_set value_types_impl(const common_chat_schema &
                // a cycle contributes no type, to be safe
                return {};
            }
-            auto types = value_types_impl(*target, visited, depth);
+            auto types = value_types_impl(*target, visited);
            visited.erase(target);
            return types;
        }
        case common_chat_schema::KIND_ANY_OF: {
            common_chat_schema::type_set types;
            for (const auto & child : static_cast<const common_chat_schema_any_of &>(s).children) {
-                types |= value_types_impl(*child, visited, depth);
+                types |= value_types_impl(*child, visited);
            }
            return types;
        }
        case common_chat_schema::KIND_ALL_OF: {
            auto types = common_chat_schema::type_set::all();
            for (const auto & child : static_cast<const common_chat_schema_all_of &>(s).children) {
-                types &= value_types_impl(*child, visited, depth);
+                types &= value_types_impl(*child, visited);
            }
            return types;
        }
@@ -444,12 +423,10 @@ static common_chat_schema::type_set value_types_impl(const common_chat_schema &

 common_chat_schema::type_set common_chat_schema::value_types() const {
    std::unordered_set<const common_chat_schema *> visited;
-    int depth = 0;
-    return value_types_impl(*this, visited, depth);
+    return value_types_impl(*this, visited);
 }

-static bool may_be_string_impl(const common_chat_schema & s, std::unordered_set<const common_chat_schema *> & visited, int & depth) {
-    depth_guard guard(depth);
+static bool may_be_string_impl(const common_chat_schema & s, std::unordered_set<const common_chat_schema *> & visited) {
    switch (s.kind()) {
        case common_chat_schema::KIND_STRING:
            return true;
@@ -468,13 +445,13 @@ static bool may_be_string_impl(const common_chat_schema & s, std::unordered_set<
            if (!target || !visited.insert(target).second) {
                return false;
            }
-            bool result = may_be_string_impl(*target, visited, depth);
+            bool result = may_be_string_impl(*target, visited);
            visited.erase(target);
            return result;
        }
        case common_chat_schema::KIND_ANY_OF:
            for (const auto & child : static_cast<const common_chat_schema_any_of &>(s).children) {
-                if (may_be_string_impl(*child, visited, depth)) {
+                if (may_be_string_impl(*child, visited)) {
                    return true;
                }
            }
@@ -486,7 +463,7 @@ static bool may_be_string_impl(const common_chat_schema & s, std::unordered_set<
                if (child->kind() == common_chat_schema::KIND_ANY) {
                    continue;
                }
-                if (!may_be_string_impl(*child, visited, depth)) {
+                if (!may_be_string_impl(*child, visited)) {
                    return false;
                }
                any_string = true;
@@ -500,8 +477,7 @@ static bool may_be_string_impl(const common_chat_schema & s, std::unordered_set<

 bool common_chat_schema::may_be_string() const {
    std::unordered_set<const common_chat_schema *> visited;
-    int depth = 0;
-    return may_be_string_impl(*this, visited, depth);
+    return may_be_string_impl(*this, visited);
 }

 const char * common_chat_schema::kind_name(node_kind kind) {
@@ -1,5 +1,4 @@
 #include "parsers.h"
-#include "log.h"

 // The DeepSeek V4 reference implementation renders consecutive tool results into a single
 // user block, ordered by the tool call order of the preceding assistant message (matched
@@ -76,30 +75,13 @@ common_chat_params common_chat_params_init_deepseek_v3_2(const common_chat_templ
    // non-thinking generation prompt ends with a bare </think> instead of an empty
    // <think></think> pair.
    const bool is_v4 = tmpl.source().find("function_calls") == std::string::npos;
-    // V4.1: DSML tag names carry a leading space, e.g. <｜DSML｜ calls>
-    const bool is_v41 = is_v4 && tmpl.source().find("｜DSML｜ calls>") != std::string::npos;

    std::optional<json> adjusted_messages;
    if (is_v4) {
        adjusted_messages = deepseek_v4_sort_tool_results(inputs.messages);
    }

-    // V4.1 can declare tools on a system message instead of top-level "tools": parse calls to those too.
-    json tools = inputs.tools;
-    if (is_v41 && !(tools.is_array() && !tools.empty()) && inputs.messages.is_array()) {
-        for (const auto & msg : inputs.messages) {
-            if (msg.value("role", "") == "system" && msg.contains("tools") && msg.at("tools").is_array()) {
-                if (!tools.is_array()) {
-                    tools = json::array();
-                }
-                for (const auto & tool : msg.at("tools")) {
-                    tools.push_back(tool);
-                }
-            }
-        }
-    }
-
-    auto has_tools           = tools.is_array() && !tools.empty();
+    auto has_tools           = inputs.tools.is_array() && !inputs.tools.empty();
    auto has_response_format = !inputs.json_schema.is_null() && inputs.json_schema.is_object();
    auto extract_reasoning   = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE;
    auto include_grammar     = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE);
@@ -108,39 +90,17 @@ common_chat_params common_chat_params_init_deepseek_v3_2(const common_chat_templ
    if (is_v4 && has_response_format) {
        additional_context = json{ { "response_format", inputs.json_schema } };
    }
-    // V4.1 raises on anything but low/high/max or 1-100: map minimal -> low, xhigh -> max, medium -> default
-    if (is_v41 && inputs.extra_context.is_object() && inputs.extra_context.contains("reasoning_effort") &&
-        inputs.extra_context.at("reasoning_effort").is_string()) {
-        const std::string effort = inputs.extra_context.at("reasoning_effort").get<std::string>();
-        json mapped;
-        if (effort == "minimal") {
-            mapped = "low";
-        } else if (effort == "xhigh") {
-            mapped = "max";
-        } else if (effort == "medium") {
-            mapped = nullptr;
-        }
-        if (effort == "minimal" || effort == "xhigh" || effort == "medium") {
-            LOG_DBG("%s: DeepSeek V4.1 reasoning_effort '%s' -> %s\n", __func__, effort.c_str(),
-                    mapped.is_null() ? "unset (template default)" : mapped.get<std::string>().c_str());
-            if (!additional_context.has_value()) {
-                additional_context = json::object();
-            }
-            (*additional_context)["reasoning_effort"] = mapped;
-        }
-    }

    const std::string DSML         = "｜DSML｜";
    const std::string THINK_START  = "<think>";
    const std::string THINK_END    = "</think>";
-    const std::string TAG_SP       = is_v41 ? " " : "";
-    const std::string TC_BLOCK     = is_v41 ? "calls" : is_v4 ? "tool_calls" : "function_calls";
-    const std::string FC_START     = "<" + DSML + TAG_SP + TC_BLOCK + ">";
-    const std::string FC_END       = "</" + DSML + TAG_SP + TC_BLOCK + ">";
-    const std::string INVOKE_START = "<" + DSML + TAG_SP + "invoke";
-    const std::string INVOKE_END   = "</" + DSML + TAG_SP + "invoke>";
-    const std::string PARAM_START  = "<" + DSML + TAG_SP + "parameter";
-    const std::string PARAM_END    = "</" + DSML + TAG_SP + "parameter>";
+    const std::string TC_BLOCK     = is_v4 ? "tool_calls" : "function_calls";
+    const std::string FC_START     = "<" + DSML + TC_BLOCK + ">";
+    const std::string FC_END       = "</" + DSML + TC_BLOCK + ">";
+    const std::string INVOKE_START = "<" + DSML + "invoke";
+    const std::string INVOKE_END   = "</" + DSML + "invoke>";
+    const std::string PARAM_START  = "<" + DSML + "parameter";
+    const std::string PARAM_END    = "</" + DSML + "parameter>";
    const std::string GEN_PROMPT   = "<｜Assistant｜>";
    const std::string TC_SEPARATOR = "\n\n";

@@ -192,7 +152,7 @@ common_chat_params common_chat_params_init_deepseek_v3_2(const common_chat_templ
        // build tool call section first since we might need it in reasoning
        auto tool_choice = p.choice();
        if (has_tool_calls) {
-            foreach_function(tools, [&](size_t tool_index, const json & tool) {
+            foreach_function(inputs.tools, [&](size_t tool_index, const json & tool) {
                const auto & function = tool.at("function");
                std::string  name     = function.at("name");

@@ -263,10 +223,7 @@ common_chat_params common_chat_params_init_deepseek_v3_2(const common_chat_templ
        }

        if (extract_reasoning && inputs.enable_thinking) {
-            // V4.1 renders a continuation without reasoning as "<｜Assistant｜></think>" even in thinking mode,
-            // so consume the bare THINK_END too or it leaks into the content
-            reasoning = is_v41 ? p.optional((THINK_START + p.reasoning(p.until(THINK_END)) + THINK_END) | p.literal(THINK_END))
-                               : p.optional(THINK_START + p.reasoning(p.until(THINK_END)) + THINK_END);
+            reasoning = p.optional(THINK_START + p.reasoning(p.until(THINK_END)) + THINK_END);
            reasoning_with_tc = THINK_START +
                p.reasoning(p.until_one_of({ TC_SEPARATOR + FC_START, FC_START, THINK_END })) +
                p.space() + obligatory_tool_calls;
@@ -76,10 +76,9 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat

        auto reasoning = p.eps();
        if (supports_reasoning && extract_reasoning) {
-            // The template strips the newline before </think>, so leave it out of the reasoning
            reasoning = p.optional("<think>" + p.space() +
-                                   p.reasoning(p.until_one_of({ "\n</think>", "</think>", "<tool_call>" })) +
-                                   (p.literal("\n</think>") | p.literal("</think>") | p.peek(p.literal("<tool_call>"))));
+                                   p.reasoning(p.until_one_of({ "</think>", "<tool_call>" })) +
+                                   (p.literal("</think>") | p.peek(p.literal("<tool_call>"))));
        }

        // Response format parser
@@ -167,15 +166,8 @@ common_chat_params common_chat_params_init_qwen3_coder(const common_chat_templat
            auto calls      = inputs.parallel_tool_calls ? tool_call_first + p.zero_or_more(tool_call) : tool_call_first;
            auto tool_calls = p.trigger_rule("tool-call-root", p.repeat(calls, min_calls, 1));

-            // The templates separate content from the tool calls with one or two newlines,
-            // stop the content before them
-            std::vector<std::string> content_ends;
-            for (const auto & start : tool_call_starts) {
-                content_ends.insert(content_ends.end(), { "\n\n" + start, "\n" + start, start });
-            }
-
            return generation_prompt +
-                   (reasoning << p.content(p.until_one_of(content_ends)) << tool_calls);
+                   (reasoning << p.content(p.until_one_of(tool_call_starts)) << tool_calls);
        }

        // Content only parser
@@ -10,12 +10,10 @@
 #include <initializer_list>
 #include <map>
 #include <memory>
+#include <regex>
 #include <set>
 #include <stdexcept>

-// The parser can run with a lower limit than COMMON_RECURSIVE_LIMIT
-static constexpr int RULE_RECURSIVE_LIMIT = 64;
-
 // Trick to catch missing branches
 template <typename T>
 inline constexpr bool is_always_false_v = false;
@@ -188,17 +186,13 @@ std::string common_peg_ast_node::sanitized_text() const {
 }

 void common_peg_ast_arena::visit(common_peg_ast_id id, const common_peg_ast_visitor & visitor) const {
-    // no recursion: the AST is several nodes deep per level of the input
-    std::vector<common_peg_ast_id> stack = { id };
-    while (!stack.empty()) {
-        const auto cur = stack.back();
-        stack.pop_back();
-        if (cur == COMMON_PEG_INVALID_AST_ID) {
-            continue;
-        }
-        const auto & node = get(cur);
-        visitor(node);
-        stack.insert(stack.end(), node.children.rbegin(), node.children.rend());
+    if (id == COMMON_PEG_INVALID_AST_ID) {
+        return;
+    }
+    const auto & node = get(id);
+    visitor(node);
+    for (const auto & child : node.children) {
+        visit(child, visitor);
    }
 }

@@ -721,9 +715,8 @@ struct parser_executor {
                return common_peg_parse_result(COMMON_PEG_PARSE_RESULT_SUCCESS, start_pos, pos, {}, std::move(invalid_utf8));
            }

-            if (match == common_trie::PARTIAL_MATCH && !ctx.is_complete()) {
-                // Found a partial match extending to end of input, return everything before it.
-                // On a complete input the delimiter can no longer form, so the bytes belong to the text
+            if (match == common_trie::PARTIAL_MATCH) {
+                // Found a partial match extending to end of input, return everything before it
                return common_peg_parse_result(COMMON_PEG_PARSE_RESULT_SUCCESS, start_pos, pos, {}, std::move(invalid_utf8));
            }

@@ -743,15 +736,8 @@ struct parser_executor {
    }

    common_peg_parse_result operator()(const common_peg_rule_parser & p) {
-        // only rules recurse, a recursive rule is entered once per input level
-        if (ctx.rule_depth[p.child] >= RULE_RECURSIVE_LIMIT) {
-            throw common_peg_recursion_error("input is nested more than " + std::to_string(RULE_RECURSIVE_LIMIT) + " levels deep");
-        }
-
        // Parse the child
-        ctx.rule_depth[p.child]++;
        auto result = arena.parse(p.child, ctx, start_pos);
-        ctx.rule_depth[p.child]--;

        if (!result.fail()) {
            std::string_view text;
@@ -833,10 +819,6 @@ common_peg_parse_result common_peg_arena::parse(common_peg_parse_context & ctx,
    if (root_ == COMMON_PEG_INVALID_PARSER_ID) {
        throw std::runtime_error("No root parser set");
    }
-
-    // Initialize rule depth for all parsers
-    ctx.rule_depth.assign(parsers_.size(), 0);
-
    return parse(root_, ctx, start);
 }

@@ -1087,6 +1069,11 @@ common_peg_parser operator|(const std::string & str, const common_peg_parser & p
    return operator|(str.c_str(), p);
 }

+static std::string rule_name(const std::string & name) {
+    static const std::regex invalid_rule_chars_re("[^a-zA-Z0-9-]+");
+    return std::regex_replace(name, invalid_rule_chars_re, "-");
+}
+
 common_peg_parser_builder::common_peg_parser_builder() {}

 common_peg_parser common_peg_parser_builder::sequence(const std::vector<common_peg_parser_id> & parsers) {
@@ -1168,14 +1155,14 @@ common_peg_parser common_peg_parser_builder::schema(const common_peg_parser & p,
 }

 common_peg_parser common_peg_parser_builder::rule(const std::string & name, const common_peg_parser & p, bool trigger) {
-    auto clean_name = gbnf_format_rule_name(name);
+    auto clean_name = rule_name(name);
    auto rule_id = arena_.add_parser(common_peg_rule_parser{clean_name, p.id(), trigger});
    arena_.add_rule(clean_name, rule_id);
    return ref(clean_name);
 }

 common_peg_parser common_peg_parser_builder::rule(const std::string & name, const std::function<common_peg_parser()> & builder_fn, bool trigger) {
-    auto clean_name = gbnf_format_rule_name(name);
+    auto clean_name = rule_name(name);
    if (arena_.has_rule(clean_name)) {
        return ref(clean_name);
    }
@@ -6,7 +6,6 @@

 #include <memory>
 #include <set>
-#include <stdexcept>
 #include <unordered_map>
 #include <string>
 #include <string_view>
@@ -162,10 +161,9 @@ struct common_peg_parse_result {
 };

 enum common_peg_parse_flags {
-    COMMON_PEG_PARSE_FLAG_NONE     = 0,
-    COMMON_PEG_PARSE_FLAG_LENIENT  = 1 << 0,
-    COMMON_PEG_PARSE_FLAG_DEBUG    = 1 << 1,
-    COMMON_PEG_PARSE_FLAG_COMPLETE = 1 << 2,
+    COMMON_PEG_PARSE_FLAG_NONE    = 0,
+    COMMON_PEG_PARSE_FLAG_LENIENT = 1 << 0,
+    COMMON_PEG_PARSE_FLAG_DEBUG   = 1 << 1,
 };

 inline common_peg_parse_flags operator|(common_peg_parse_flags a, common_peg_parse_flags b) {
@@ -184,10 +182,6 @@ inline common_peg_parse_flags operator~(common_peg_parse_flags a) {
    return static_cast<common_peg_parse_flags>(~int(a));
 }

-struct common_peg_recursion_error : std::invalid_argument {
-    using std::invalid_argument::invalid_argument;
-};
-
 struct common_peg_parse_context {
    std::string input;               // [h,   e,  l,  l,  o,  _,  w,  o,  r,  l,  d]
    std::vector<llama_token> tokens; // [id, -1, -1, -1, -1, id, -1, -1, -1, -1, -1]
@@ -195,7 +189,6 @@ struct common_peg_parse_context {
    common_peg_ast_arena ast;

    int parse_depth;
-    std::vector<int> rule_depth; // index: rule child id

    common_peg_parse_context(common_peg_parse_flags flags = COMMON_PEG_PARSE_FLAG_NONE)
        : flags(flags), parse_depth(0) {}
@@ -210,7 +203,6 @@ struct common_peg_parse_context {

    bool is_lenient() const { return flags & COMMON_PEG_PARSE_FLAG_LENIENT; }
    bool is_debug() const { return flags & COMMON_PEG_PARSE_FLAG_DEBUG; }
-    bool is_complete() const { return flags & COMMON_PEG_PARSE_FLAG_COMPLETE; }
 };

 class common_peg_arena;
@@ -171,12 +171,12 @@ static std::map<std::string, std::map<std::string, std::string>> parse_ini_from_
    std::map<std::string, std::map<std::string, std::string>> parsed;

    if (!std::filesystem::exists(path)) {
-        throw std::runtime_error("preset file does not exist: " + path.string());
+        throw std::runtime_error("preset file does not exist: " + fs_path_to_utf8(path));
    }

    std::ifstream file(path);
    if (!file.good()) {
-        throw std::runtime_error("failed to open server preset file: " + path.string());
+        throw std::runtime_error("failed to open server preset file: " + fs_path_to_utf8(path));
    }

    std::string contents((std::istreambuf_iterator<char>(file)), std::istreambuf_iterator<char>());
@@ -225,7 +225,7 @@ static std::map<std::string, std::map<std::string, std::string>> parse_ini_from_
    common_peg_parse_context ctx(contents);
    const auto result = parser.parse(ctx);
    if (!result.success()) {
-        throw std::runtime_error("failed to parse server config file: " + path.string());
+        throw std::runtime_error("failed to parse server config file: " + fs_path_to_utf8(path));
    }

    std::string current_section = COMMON_PRESET_DEFAULT_NAME;
@@ -323,7 +323,7 @@ common_presets common_preset_context::load_from_ini(const std::filesystem::path
                }
                LOG_DBG("accepted option: %s = %s\n", key.c_str(), preset.options[opt].c_str());
            } else if (ignore_unknown_keys) {
-                LOG_WRN("ignoring option '%s' from %s: not supported by this program\n", key.c_str(), path.string().c_str());
+                LOG_WRN("ignoring option '%s' from %s: not supported by this program\n", key.c_str(), fs_path_to_utf8(path).c_str());
            } else {
                throw std::runtime_error(string_format(
                    "option '%s' not recognized in preset '%s'",
@@ -401,7 +401,7 @@ common_presets common_preset_context::load_from_models_dir(const std::string & m
            if (!entry.is_regular_file(ec)) {
                continue;
            }
-            const std::string fname = entry.path().filename().string();
+            const std::string fname = fs_path_to_utf8(entry.path().filename());
            if (!string_ends_with(fname, ".gguf")) {
                continue;
            }
@@ -422,9 +422,9 @@ common_presets common_preset_context::load_from_models_dir(const std::string & m
        if (!path.empty()) {
            models.push_back({
                /* name        */ name,
-                /* path        */ path.string(),
-                /* path_mmproj */ mmproj_file.string(), // can be empty
-                /* path_draft  */ draft_file.string()   // can be empty
+                /* path        */ fs_path_to_utf8(path),
+                /* path_mmproj */ fs_path_to_utf8(mmproj_file), // can be empty
+                /* path_draft  */ fs_path_to_utf8(draft_file)    // can be empty
            });
        }
    };
@@ -432,13 +432,13 @@ common_presets common_preset_context::load_from_models_dir(const std::string & m
    for (const auto & entry : std::filesystem::directory_iterator(dir)) {
        std::error_code ec;
        if (entry.is_directory(ec)) {
-            scan_subdir(entry.path(), entry.path().filename().string());
+            scan_subdir(entry.path(), fs_path_to_utf8(entry.path().filename()));
            continue;
        }
        if (!entry.is_regular_file(ec)) {
            continue;
        }
-        const std::string fname = entry.path().filename().string();
+        const std::string fname = fs_path_to_utf8(entry.path().filename());
        if (!string_ends_with(fname, ".gguf")) {
            continue;
        }
@@ -450,7 +450,7 @@ common_presets common_preset_context::load_from_models_dir(const std::string & m
        string_replace_all(name, ".gguf", "");
        models.push_back({
            /* name        */ name,
-            /* path        */ entry.path().string(),
+            /* path        */ fs_path_to_utf8(entry.path()),
            /* path_mmproj */ "",
            /* path_draft  */ ""
        });
@@ -136,8 +136,7 @@ static void common_reasoning_budget_accept(struct llama_sampler * smpl, llama_to
            // track the end sequence within forced_tokens so it is also reported on DONE
            const int32_t match = ctx->end_matcher.advance(token);
            ctx->force_pos++;
-            // an end tag that is already in the prompt (e.g. "<think></think>") also ends forcing
-            if (match >= 0 || ctx->force_pos >= ctx->forced_tokens.size()) {
+            if (ctx->force_pos >= ctx->forced_tokens.size()) {
                ctx->state = REASONING_BUDGET_DONE;
                ctx->end_match = match;
                COM_TRC("%s", "forced sequence complete, done\n");
@@ -19,7 +19,6 @@
 #include <fstream>
 #include <iomanip>
 #include <map>
-#include <random>
 #include <cinttypes>

 #define SPC_DBG(fmt, ...) LOG_DBG("spec %12.*s: " fmt, 12, __func__, __VA_ARGS__)
@@ -1011,10 +1010,6 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
    bool    is_mrope       = false;
    int32_t selector_top_k = 0;

-    // DFlash2 probabilistic drafting: one generator per sequence and the seed it was last seeded with (-1: reseed)
-    std::vector<std::mt19937> dflash2_rng;
-    std::vector<int64_t>      dflash2_seed;
-
    // draft-dspark: the draft carries a Markov head and uses an anchor-first block layout
    const bool is_dspark;

@@ -1065,8 +1060,6 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {

        selector_top_k = llama_model_dflash_selector_top_k(model_dft);
        is_dflash2     = selector_top_k > 0;
-        dflash2_rng.resize(n_seq);
-        dflash2_seed.assign(n_seq, -1);
        char lilicorr_meta[32] = {};
        lilicorr_top_k = llama_model_meta_val_str(model_dft, "lilicorr.candidate_topk", lilicorr_meta, sizeof(lilicorr_meta)) >= 0
                ? std::atoi(lilicorr_meta) : 0;
@@ -1165,9 +1158,6 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
            return;
        }

-        // reseed per request rather than per round, or two identical requests differ
-        dflash2_seed[seq_id] = -1;
-
        const int32_t N = (int32_t) prompt.size();
        if (N <= 0) {
            return;
@@ -1348,21 +1338,6 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
                const float * lattice = llama_get_embeddings_nextn(ctx_dft);
                GGML_ASSERT(lattice && "DFlash2 selector produced no lattice");

-                // DFlash2 can instead sample each step and hand q to the verifier for rejection sampling
-                const bool sample = is_dflash2 && params.probabilistic && dp.result_q;
-
-                // greedy keeps temperature 1, sampling uses the target's
-                const float temp = sample ? dp.temp : 1.0f;
-
-                if (sample && dflash2_seed[seq_id] != dp.seed) {
-                    dflash2_seed[seq_id] = dp.seed;
-                    // the default seed means random; mixed so it differs from the target's, as in spec_retune
-                    dflash2_rng[seq_id].seed(dp.seed == LLAMA_DEFAULT_SEED ? std::random_device{}() : dp.seed ^ 0x85ebca6bu);
-                }
-
-                std::vector<float>   probs;
-                std::vector<int32_t> sorted_idx;
-
                int32_t predecessor = 0;
                for (int32_t i = 1; i < (is_dflash2 ? n_block_tokens : params.n_max + 1); ++i) {
                    const float * row = lattice + (size_t) (beg + i) * n_embd_dec;
@@ -1371,53 +1346,21 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {

                    predecessor = (int32_t) std::distance(scores,
                            std::max_element(scores, scores + top_k));
-                    if (params.p_min > 0.0f || sample) {
-                        // softmax(scores / temp) at the argmax, i.e. 1 / sum(exp((s_k - s_max) / temp))
-                        probs.resize(top_k);
+                    if (params.p_min > 0.0f) {
+                        // softmax(scores) at the argmax, i.e. 1 / sum(exp(s_k - s_max))
                        float sum = 0.0f;
                        for (int32_t k = 0; k < top_k; ++k) {
-                            probs[k] = std::exp((scores[k] - scores[predecessor]) / temp);
-                            sum += probs[k];
+                            sum += std::exp(scores[k] - scores[predecessor]);
                        }
                        if (1.0f / sum < params.p_min) {
                            break;
                        }
-                        for (int32_t k = 0; k < top_k; ++k) {
-                            probs[k] /= sum;
-                        }
-                    }
-                    if (sample) {
-                        // ggml_top_k leaves candidates unordered, so walk them by token id to keep a seed's draft stable
-                        sorted_idx.resize(top_k);
-                        for (int32_t k = 0; k < top_k; ++k) {
-                            sorted_idx[k] = k;
-                        }
-                        std::sort(sorted_idx.begin(), sorted_idx.end(), [row](int32_t a, int32_t b) { return row[a] < row[b]; });
-
-                        // a round-off overshoot keeps the argmax, which has q > 0
-                        double u = std::uniform_real_distribution<double>(0.0, 1.0)(dflash2_rng[seq_id]);
-                        for (const int32_t k : sorted_idx) {
-                            u -= probs[k];
-                            if (u < 0.0) {
-                                predecessor = k;
-                                break;
-                            }
-                        }
-
-                        auto & q = dp.result_q->emplace_back();
-                        q.reserve(top_k);
-                        for (int32_t k = 0; k < top_k; ++k) {
-                            q.push_back({ (llama_token) row[k], scores[k], probs[k] });
-                        }
                    }
                    result.push_back((llama_token) row[predecessor]);
                }

                if (result.size() < (size_t) params.n_min) {
                    result.clear();
-                    if (sample) {
-                        dp.result_q->clear();
-                    }
                }
                continue;
            }
@@ -73,7 +73,7 @@ struct common_speculative_draft_params {
    // the generated draft from the last _draft() call
    llama_tokens * result;

-    // candidate distribution per drafted token; set it to make draft-simple, draft-mtp and draft-dflash (DFlash2 only) sample
+    // candidate distribution per drafted token; set it to make draft-simple and draft-mtp sample
    std::vector<std::vector<llama_token_data>> * result_q = nullptr;

    // the target's temp and seed, read only when the drafter samples probabilistically
@@ -1,9 +0,0 @@
-<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
-<assembly xmlns="urn:schemas-microsoft-com:asm.v1" manifestVersion="1.0">
-  <application xmlns="urn:schemas-microsoft-com:asm.v3">
-    <windowsSettings>
-      <activeCodePage xmlns="http://schemas.microsoft.com/SMI/2019/WindowsSettings">UTF-8</activeCodePage>
-      <longPathAware xmlns="http://schemas.microsoft.com/SMI/2016/WindowsSettings">true</longPathAware>
-    </windowsSettings>
-  </application>
-</assembly>
@@ -17,10 +17,8 @@ __all__ = [


 TEXT_MODEL_MAP: dict[str, str] = {
-    "NEOChatModel": "sensenova_u1",
    "AfmoeForCausalLM": "afmoe",
    "LagunaForCausalLM": "laguna",
-    "FlintForCausalLM": "flint",
    "ApertusForCausalLM": "llama",
    "ArceeForCausalLM": "llama",
    "ArcticForCausalLM": "arctic",
@@ -70,7 +68,6 @@ TEXT_MODEL_MAP: dict[str, str] = {
    "LingDSparkModel": "qwen",
    "DeepseekV4ForCausalLM": "deepseek",
    "DeepseekV4DSparkModel": "deepseek",
-    "DeepseekV41ForCausalLM": "deepseek41",
    "DistilBertForMaskedLM": "bert",
    "DistilBertForSequenceClassification": "bert",
    "DistilBertModel": "bert",
@@ -172,11 +169,6 @@ TEXT_MODEL_MAP: dict[str, str] = {
    "D1OmniModel": "lfm2",
    "Lfm2Model": "lfm2",
    "Lfm2MoeForCausalLM": "lfm2",
-    "Lfm3D1Model": "lfm3",
-    "Lfm3ForCausalLM": "lfm3",
-    "Lfm3OmniForConditionalGeneration": "lfm3",
-    "Lfm3VlForConditionalGeneration": "lfm3",
-    "LimiteForCausalLM": "limite",
    "Llama4ForCausalLM": "llama",
    "Llama4ForConditionalGeneration": "llama",
    "LlamaBidirectionalModel": "llama",
@@ -202,8 +194,6 @@ TEXT_MODEL_MAP: dict[str, str] = {
    "MiniCPMForCausalLM": "minicpm",
    "MiniCPMV4_6ForConditionalGeneration": "minicpm",
    "MiniCPMV4_7ForConditionalGeneration": "minicpm",
-    "MiniCPMOCRForSequenceClassification": "qwen",
-    "MiniCPMOcrForConditionalGeneration": "minicpm",
    "MiniMaxText01ForCausalLM": "minimax",
    "MiniMaxM1ForCausalLM": "minimax",
    "MiniMaxM2ForCausalLM": "minimax",
@@ -213,7 +203,6 @@ TEXT_MODEL_MAP: dict[str, str] = {
    "Ministral3Model": "mistral3",
    "Mistral3ForConditionalGeneration": "mistral3",
    "Mistral3Model": "mistral3",
-    "MistralLarge4ForConditionalGeneration": "mistral4",
    "VoxtralRealtimeForConditionalGeneration": "mistral3",
    "MistralForCausalLM": "llama",
    "MixtralForCausalLM": "llama",
@@ -294,7 +283,6 @@ TEXT_MODEL_MAP: dict[str, str] = {
    "SeedOssForCausalLM": "olmo",
    "SmallThinkerForCausalLM": "smallthinker",
    "SmolLM3ForCausalLM": "llama",
-    "SopranoTextModel": "soprano",
    "Spark2_5ForCausalLM": "spark2_5",
    "SolarOpenForCausalLM": "glm",
    "StableLMEpochForCausalLM": "stablelm",
@@ -334,7 +322,6 @@ MMPROJ_MODEL_MAP: dict[str, str] = {
    "VllmD3Model": "vllm_d3",
    "DeepseekOCR2ForCausalLM": "deepseek",
    "DeepseekOCRForCausalLM": "deepseek",
-    "DeepseekV41ForCausalLM": "deepseek41",
    "DeepseekV4ForCausalLM": "deepseek",
    "Dots3NoteForCausalLM": "dots3",
    "Dots3NoteForConditionalGeneration": "dots3",
@@ -364,9 +351,6 @@ MMPROJ_MODEL_MAP: dict[str, str] = {
    "Lfm2AudioForConditionalGeneration": "lfm2",
    "D1OmniModel": "lfm2",
    "Lfm2VlForConditionalGeneration": "lfm2",
-    "Lfm3AudioForConditionalGeneration": "lfm3",
-    "Lfm3OmniForConditionalGeneration": "lfm3",
-    "Lfm3VlForConditionalGeneration": "lfm3",
    "LightOnOCRForConditionalGeneration": "lighton_ocr",
    "Llama4ForConditionalGeneration": "llama4",
    "LlavaForConditionalGeneration": "llava",
@@ -375,8 +359,6 @@ MMPROJ_MODEL_MAP: dict[str, str] = {
    "MiniMaxM3SparseForConditionalGeneration": "minimax",
    "MiniCPMV4_6ForConditionalGeneration": "minicpm",
    "MiniCPMV4_7ForConditionalGeneration": "minicpm",
-    "MiniCPMOCRForSequenceClassification": "qwen3vl",
-    "MiniCPMOcrForConditionalGeneration": "minicpm",
    "Mistral3ForConditionalGeneration": "llava",
    "Mistral3Model": "llava",
    "NemotronH_Omni_Reasoning_V3": "nemotron",
@@ -385,7 +367,6 @@ MMPROJ_MODEL_MAP: dict[str, str] = {
    "MuseSparkForConditionalGeneration": "muse_spark",
    "PaddleOCRVisionModel": "ernie",
    "Phi4ForCausalLMV": "phi",
-    "NEOVisionModel": "sensenova_u1",
    "Qwen2AudioForConditionalGeneration": "ultravox",
    "Qwen2VLForConditionalGeneration": "qwenvl",
    "Qwen2VLModel": "qwenvl",
@@ -394,7 +375,6 @@ MMPROJ_MODEL_MAP: dict[str, str] = {
    "Qwen3ASRForConditionalGeneration": "qwen3vl",
    "Qwen3OmniMoeForConditionalGeneration": "qwen3vl",
    "PocketTTSModel": "pockettts",
-    "SopranoModel": "soprano",
    "Qwen3TTSForConditionalGeneration": "qwen3tts",
    "Qwen3VLForConditionalGeneration": "qwen3vl",
    "Qwen3VLMoeForConditionalGeneration": "qwen3vl",
@@ -16,7 +16,6 @@ from pathlib import Path
 from hashlib import sha256
 from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Literal, Sequence, TypeVar, cast
 from itertools import chain
-from math import prod
 from transformers import AutoConfig

 import numpy as np
@@ -156,7 +155,6 @@ class ModelBase:
        self.fuse_gate_up_exps = fuse_gate_up_exps
        self._gate_exp_buffer: dict[int, Tensor] = {}
        self._up_exp_buffer: dict[int, Tensor] = {}
-        self._unfusable_gate_up_layers: set[int] = set()
        self.fuse_qkv = fuse_qkv
        self._q_buffer: dict[int, Tensor] = {}
        self._k_buffer: dict[int, Tensor] = {}
@@ -177,9 +175,7 @@ class ModelBase:
        self._nvfp4_layer_algo: dict[str, str | None] = {} # per-layer quant_algo, keyed by HF module path
        self._prec_a4: dict[str, bool] = {} # gguf tensor name -> can use 4-bit (A4) activations
        self._fp8_as_q8 = fp8_as_q8
-        self._preserve_fp8 = ftype == gguf.LlamaFileType.GUESSED and not fp8_as_q8
        self._fp8_dequantized: set[str] = set()
-        self._fp8_e4m3_preserved: set[str] = set()

        # Apply heuristics to figure out typical tensor encoding based on first tensor's dtype
        # NOTE: can't use field "torch_dtype" in config.json, because some finetunes lie.
@@ -330,7 +326,7 @@ class ModelBase:

    def dequant_model(self):
        # If all quantized tensors were already handled (e.g. pure NVFP4), skip
-        if self._is_nvfp4 and not any(k.endswith((".weight_scale", ".weight_scale_inv", ".input_scale", ".activation_scale", "_activation_scale", ".k_scale", ".v_scale")) for k in self.model_tensors):
+        if self._is_nvfp4 and not any(k.endswith((".weight_scale", ".weight_scale_inv")) for k in self.model_tensors):
            return

        tensors_to_remove: list[str] = []
@@ -354,12 +350,6 @@ class ModelBase:
            def dequant_simple(weight: Tensor, scale: Tensor, block_size: Sequence[int] | None = None) -> Tensor:
                scale = scale.float()

-                # ModelOpt block scales are pre-shaped for broadcasting, e.g. [out/128, 1, in/128, 1]
-                if block_size is None and scale.ndim == 2 * weight.ndim and all(n == 1 for n in scale.shape[1::2]):
-                    assert all(w % n == 0 for w, n in zip(weight.shape, scale.shape[0::2]))
-                    block_size = [w // n for w, n in zip(weight.shape, scale.shape[0::2])]
-                    scale = scale.reshape(tuple(scale.shape[0::2]))
-
                if block_size is not None:
                    dim_offset = scale.ndim - len(block_size)
                    for i, size in enumerate(block_size):
@@ -450,7 +440,7 @@ class ModelBase:

                return (unpacked * scale.unsqueeze(-1).float()).reshape(shape)

-            def dequant_fp8(fp8_configs: list[dict[str, Any]] | None = None) -> None:
+            def dequant_fp8() -> None:
                for name in self.model_tensors.keys():
                    if name.endswith(".weight_scale"):
                        weight_name = name.removesuffix("_scale")
@@ -459,29 +449,14 @@ class ModelBase:
                            continue
                        w = self.model_tensors[weight_name]
                        s = self.model_tensors[name]
-                        block_size = None
                        is_fp8_weight = False
-                        if self._fp8_as_q8 or fp8_configs is not None:
-                            weight = w()
-                            is_fp8_weight = weight.dtype in (torch.float8_e4m3fn, torch.float8_e5m2)
-                        if fp8_configs is not None:
-                            if not is_fp8_weight:
-                                raise ValueError(f"Expected remaining weight {weight_name!r} to be FP8, got {weight.dtype}")
-                            scale = s()
-                            is_channel_scale = scale.ndim >= 1 and scale.ndim <= weight.ndim and scale.shape[0] == weight.shape[0] and all(size == 1 for size in scale.shape[1:])
-                            if prod(scale.shape) != 1 and not is_channel_scale:
-                                block_configs = [c for c in fp8_configs if c.get("strategy") == "block"]
-                                if not block_configs or any(not c.get("block_structure") or c.get("group_size") is not None for c in block_configs):
-                                    raise ValueError(f"FP8 weight {weight_name!r} with scale shape {list(scale.shape)} requires a block layout")
-                                block_sizes = {tuple(c["block_structure"]) for c in block_configs}
-                                if len(block_sizes) != 1:
-                                    raise ValueError(f"FP8 weight {weight_name!r} has ambiguous block layouts")
-                                block_size = next(iter(block_sizes))
-                        self.model_tensors[weight_name] = lambda w=w, s=s, bs=block_size: dequant_simple(w(), s(), bs)
+                        if self._fp8_as_q8:
+                            is_fp8_weight = w().dtype in (torch.float8_e4m3fn, torch.float8_e5m2)
+                        self.model_tensors[weight_name] = lambda w=w, s=s: dequant_simple(w(), s(), None)
                        tensors_to_remove.append(name)
-                        if self._fp8_as_q8 and is_fp8_weight:
+                        if is_fp8_weight:
                            self._fp8_dequantized.add(weight_name)
-                    if name.endswith((".input_scale", ".activation_scale", "_activation_scale", ".k_scale", ".v_scale")):
+                    if name.endswith((".input_scale", ".k_scale", ".v_scale")):
                        tensors_to_remove.append(name)

            if quant_method == "bitnet":
@@ -546,37 +521,29 @@ class ModelBase:
                nvfp4_compressed_tensors = self._is_nvfp4_compressed_tensors(quant_method, quant_format, groups)

                if nvfp4_compressed_tensors:
-                    fp8_configs = [
-                        g["weights"] for g in groups.values()
-                        if isinstance(g, dict) and g.get("format") == "float-quantized"
-                        and g.get("weights", {}).get("type") == "float"
-                        and g.get("weights", {}).get("num_bits") == 8
-                    ]
-                    dequant_fp8(fp8_configs)
+                    dequant_fp8()
                elif len(groups) > 1:
                    raise NotImplementedError("Can't handle multiple config groups for compressed-tensors yet")
                elif quant_format == "float-quantized" or quant_format == "int-quantized" or quant_format == "naive-quantized":
                    weight_config = tuple(groups.values())[0]["weights"]
                    block_size = weight_config.get("block_structure", None)
                    strategy = weight_config.get("strategy")
+                    assert strategy == "channel" or strategy == "block"
+                    assert weight_config.get("group_size") is None  # didn't find a model using this yet
                    is_fp8 = (
                        quant_format == "float-quantized"
                        and weight_config.get("type") == "float"
                        and weight_config.get("num_bits") == 8
                    )
-                    assert strategy in ("channel", "block") or (is_fp8 and strategy == "tensor")
-                    assert weight_config.get("group_size") is None  # didn't find a model using this yet
                    for name in self.model_tensors.keys():
                        if name.endswith(".weight_scale"):
                            weight_name = name.removesuffix("_scale")
                            w = self.model_tensors[weight_name]
                            s = self.model_tensors[name]
-                            self.model_tensors[weight_name] = lambda w=w, s=s, bs=block_size: dequant_simple(w(), s(), bs)
+                            self.model_tensors[weight_name] = lambda w=w, s=s: dequant_simple(w(), s(), block_size)
                            tensors_to_remove.append(name)
                            if self._fp8_as_q8 and is_fp8:
                                self._fp8_dequantized.add(weight_name)
-                        if is_fp8 and name.endswith((".input_scale", ".activation_scale", "_activation_scale", ".k_scale", ".v_scale")):
-                            tensors_to_remove.append(name)
                elif quant_format == "pack-quantized":
                    weight_config = tuple(groups.values())[0]["weights"]
                    assert weight_config.get("strategy") == "group"
@@ -614,148 +581,6 @@ class ModelBase:
        for name, value in new_tensors.items():
            self.model_tensors[name] = value

-    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
-        new_name = self.tensor_map.get_name(name, try_suffixes=(".weight",))
-        return (new_name,) if new_name is not None else ()
-
-    def _prepare_fp8_e4m3_tensors(self):
-        if not self._preserve_fp8:
-            return
-
-        scale_tensors: dict[str, list[tuple[int, float]] | np.ndarray] = {}
-        input_scale_tensors: dict[str, list[tuple[int, float]] | np.ndarray] = {}
-        consumed: list[str] = []
-        candidates: dict[str, tuple[str, Tensor, tuple[str, ...], torch.Size]] = {}
-        weight_dtypes: dict[str, torch.dtype] = {}
-        scaled_weights: set[str] = set()
-        expert_pattern = re.compile(r"(\.[^.]*experts)\.(\d+)\.")
-        scale_types = {
-            gguf.MODEL_TENSOR.ATTN_Q, gguf.MODEL_TENSOR.ATTN_K, gguf.MODEL_TENSOR.ATTN_V,
-            gguf.MODEL_TENSOR.ATTN_OUT, gguf.MODEL_TENSOR.ATTN_QKV, gguf.MODEL_TENSOR.ATTN_GATE,
-            gguf.MODEL_TENSOR.FFN_GATE, gguf.MODEL_TENSOR.FFN_UP, gguf.MODEL_TENSOR.FFN_DOWN,
-            gguf.MODEL_TENSOR.FFN_GATE_SHEXP, gguf.MODEL_TENSOR.FFN_UP_SHEXP, gguf.MODEL_TENSOR.FFN_DOWN_SHEXP,
-            gguf.MODEL_TENSOR.FFN_GATE_EXP, gguf.MODEL_TENSOR.FFN_UP_EXP, gguf.MODEL_TENSOR.FFN_DOWN_EXP,
-            gguf.MODEL_TENSOR.SSM_IN, gguf.MODEL_TENSOR.SSM_OUT, gguf.MODEL_TENSOR.SSM_ALPHA, gguf.MODEL_TENSOR.SSM_BETA,
-            gguf.MODEL_TENSOR.NEXTN_EH_PROJ, gguf.MODEL_TENSOR.NEXTN_SHARED_HEAD_HEAD, gguf.MODEL_TENSOR.OUTPUT,
-        }
-
-        for scale_name in list(self.model_tensors.keys()):
-            weight_name = None
-            if scale_name.endswith(".weight_scale"):
-                weight_name = scale_name.removesuffix("_scale")
-            elif scale_name.endswith("_scale_inv"):
-                weight_name = scale_name.removesuffix("_scale_inv")
-            elif scale_name.endswith(".qscale_weight"):
-                weight_name = scale_name.removesuffix("qscale_weight") + "weight"
-
-            if weight_name is None or weight_name not in self.model_tensors:
-                continue
-
-            scaled_weights.add(weight_name)
-            weight = self.model_tensors[weight_name]()
-            weight_dtypes[weight_name] = weight.dtype
-            if weight.dtype != torch.float8_e4m3fn or weight.ndim < 2:
-                continue
-
-            scale = self.model_tensors[scale_name]()
-            if prod(scale.shape) != 1:
-                continue
-
-            merged_name = expert_pattern.sub(r"\1.", weight_name)
-            new_names = self._map_fp8_weight_names(merged_name)
-            if not new_names:
-                continue
-            if any((mapped := self.tensor_map.get_type_and_name(new_name, try_suffixes=(".weight",))) is None or mapped[0] not in scale_types for new_name in new_names):
-                continue
-            bid = next((int(part) for part in new_names[0].split(".") if part.isdecimal()), None)
-            self._fp8_e4m3_preserved.add(weight_name)
-            if any(new_name.endswith("_norm.weight") or self.tensor_force_quant(weight_name, new_name, bid, weight.ndim) != gguf.GGMLQuantizationType.F8_E4M3 for new_name in new_names):
-                self._fp8_e4m3_preserved.discard(weight_name)
-                continue
-            candidates[weight_name] = (scale_name, scale, new_names, weight.shape)
-
-        expert_groups: dict[str, list[str]] = {}
-        for name in self.model_tensors:
-            if (match := expert_pattern.search(name)) and name.endswith(".weight"):
-                expert_groups.setdefault(name[:match.end(1)], []).append(name)
-        for names in expert_groups.values():
-            for name in names:
-                if name not in weight_dtypes:
-                    weight_dtypes[name] = self.model_tensors[name]().dtype
-            dtypes = {weight_dtypes[name] for name in names}
-            if torch.float8_e4m3fn not in dtypes:
-                continue
-            if dtypes == {torch.float8_e4m3fn} and not any(name in scaled_weights for name in names):
-                continue
-            # The converter merges expert projections together.
-            if not all(name in candidates for name in names):
-                for name in names:
-                    candidates.pop(name, None)
-                    self._fp8_e4m3_preserved.discard(name)
-                    if name not in scaled_weights:
-                        gen = self.model_tensors[name]
-                        self.model_tensors[name] = lambda gen=gen: gen().float()
-
-        for weight_name, (scale_name, scale, new_names, shape) in candidates.items():
-            scale = LazyTorchTensor.to_eager(scale).float().flatten()
-
-            weight_prefix = weight_name.removesuffix(".weight")
-            # Transformers fine-grained FP8 uses activation_scale while ModelOpt uses input_scale.
-            # Ref: https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/mistral3.py#L357-L361
-            input_scale_name = next((name for name in (
-                weight_prefix + ".input_scale",
-                weight_prefix + ".activation_scale",
-                weight_prefix + "_activation_scale",
-                weight_prefix + ".qscale_act",
-            ) if name in self.model_tensors), None)
-            input_scale = None
-            if input_scale_name is not None:
-                input_scale = self.model_tensors[input_scale_name]()
-                if prod(input_scale.shape) != 1:
-                    raise ValueError(f"FP8 input scale {input_scale_name!r} must be a scalar")
-                input_scale = LazyTorchTensor.to_eager(input_scale).float().flatten()
-                consumed.append(input_scale_name)
-
-            consumed.append(scale_name)
-
-            expert_match = expert_pattern.search(weight_name)
-            if expert_match:
-                expert_id = int(expert_match.group(2))
-                for new_name in new_names:
-                    target_name = new_name.replace(".weight", ".scale")
-                    entries = scale_tensors.setdefault(target_name, [])
-                    assert isinstance(entries, list)
-                    cast(list[tuple[int, float]], entries).append((expert_id, float(scale[0])))
-                    if input_scale is not None:
-                        target_name = new_name.replace(".weight", ".input_scale")
-                        entries = input_scale_tensors.setdefault(target_name, [])
-                        assert isinstance(entries, list)
-                        cast(list[tuple[int, float]], entries).append((expert_id, float(input_scale[0])))
-            else:
-                for new_name in new_names:
-                    mapped = self.tensor_map.get_type_and_name(new_name, try_suffixes=(".weight",)) if len(shape) == 3 else None
-                    is_packed_expert = mapped is not None and mapped[0] in (
-                        gguf.MODEL_TENSOR.FFN_GATE_EXP, gguf.MODEL_TENSOR.FFN_UP_EXP, gguf.MODEL_TENSOR.FFN_DOWN_EXP,
-                    )
-                    # Packed expert weights need one scale value per expert.
-                    n_scales = shape[0] if is_packed_expert else 1
-                    scale_tensors[new_name.replace(".weight", ".scale")] = np.repeat(scale.numpy(), n_scales)
-                    if input_scale is not None:
-                        input_scale_tensors[new_name.replace(".weight", ".input_scale")] = np.repeat(input_scale.numpy(), n_scales)
-
-        for name in consumed:
-            self.model_tensors.pop(name, None)
-
-        for name, values in chain(scale_tensors.items(), input_scale_tensors.items()):
-            if isinstance(values, list):
-                values.sort(key=lambda item: item[0])
-                scale = np.array([item[1] for item in values], dtype=np.float32)
-            else:
-                scale = values.astype(np.float32)
-            if not np.allclose(scale, 1.0, atol=1e-6):
-                logger.info(f"  + {name} (FP8 scale, shape [{scale.size}])")
-                self.gguf_writer.add_tensor(name, scale)
-
    @classmethod
    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
        name, gen = item
@@ -800,26 +625,6 @@ class ModelBase:
            raise ValueError(f"Can not map tensor {name!r}")
        return new_name

-    def prepare_gate_up_fusion(self) -> None:
-        self._unfusable_gate_up_layers.clear()
-        if not self.fuse_gate_up_exps or self._fp8_as_q8:
-            return
-
-        for name, gen in self.model_tensors.items():
-            merged_name = re.sub(r"(\.experts)\.\d+\.", r"\1.", name)
-            mapped = self.tensor_map.get_type_and_name(merged_name, try_suffixes=(".weight",))
-            if mapped is None:
-                continue
-            tensor_type, new_name = mapped
-            if tensor_type not in (gguf.MODEL_TENSOR.FFN_GATE_EXP, gguf.MODEL_TENSOR.FFN_UP_EXP, gguf.MODEL_TENSOR.FFN_GATE_UP_EXP) or not new_name.endswith(".weight"):
-                continue
-            if gen().dtype == torch.float8_e4m3fn:
-                bid = next(int(part) for part in new_name.split(".") if part.isdecimal())
-                self._unfusable_gate_up_layers.add(bid)
-
-        for bid in sorted(self._unfusable_gate_up_layers):
-            logger.warning(f"Skipping --fuse-gate-up-exps for layer {bid}: fusion of preserved FP8 expert projections is not supported")
-
    def prepare_qkv_fusion(self) -> None:
        self._fusable_qkv_weight_layers.clear()
        self._fusable_qkv_bias_layers.clear()
@@ -833,7 +638,6 @@ class ModelBase:
        }
        weights: dict[int, set[gguf.MODEL_TENSOR]] = {}
        biases: dict[int, set[gguf.MODEL_TENSOR]] = {}
-        fp8_layers: set[int] = set()

        for name in self.model_tensors:
            mapped = self.tensor_map.get_type_and_name(name, try_suffixes=(".weight", ".bias"))
@@ -848,19 +652,12 @@ class ModelBase:
                continue
            if new_name.endswith(".weight"):
                weights.setdefault(bid, set()).add(tensor_type)
-                if not self._fp8_as_q8 and self.model_tensors[name]().dtype == torch.float8_e4m3fn:
-                    fp8_layers.add(bid)
            elif new_name.endswith(".bias"):
                biases.setdefault(bid, set()).add(tensor_type)

        for bid, weight_types in weights.items():
            bias_types = biases.get(bid, set())
            if weight_types == qkv_types and (not bias_types or bias_types == qkv_types):
-                # NVFP4 weights use separate reordering and repacking hooks and are exported before this fusion pass.
-                # Separate FP8 scales cannot be represented by one fused QKV scale.
-                if bid in fp8_layers:
-                    logger.warning(f"Skipping --fuse-qkv for layer {bid}: fusion of preserved FP8 projections is not supported")
-                    continue
                self._fusable_qkv_weight_layers.add(bid)
                if bias_types:
                    self._fusable_qkv_bias_layers.add(bid)
@@ -1049,8 +846,7 @@ class ModelBase:
        new_name = self.map_tensor_name(name)

        # Handle gate/up expert tensor fusion if enabled
-        # Preserved FP8 gate/up scales must stay with their separate projections.
-        if self.fuse_gate_up_exps and bid is not None and bid not in self._unfusable_gate_up_layers:
+        if self.fuse_gate_up_exps and bid is not None:
            if self.match_model_tensor_name(new_name, gguf.MODEL_TENSOR.FFN_GATE_EXP, bid):
                self._gate_exp_buffer[bid] = data_torch
            elif self.match_model_tensor_name(new_name, gguf.MODEL_TENSOR.FFN_UP_EXP, bid):
@@ -1107,46 +903,8 @@ class ModelBase:

        return [(new_name, data_torch)]

-    def _tensor_requires_f32(self, new_name: str, bid: int | None, n_dims: int) -> bool:
-        # Conditions should closely match those in llama_model_quantize_internal in llama.cpp.
-        return n_dims <= 1 or new_name.endswith("_norm.weight") or (
-            any(
-                self.match_model_tensor_name(new_name, key, bid)
-                for key in (
-                    gguf.MODEL_TENSOR.FFN_GATE_INP,
-                    gguf.MODEL_TENSOR.FFN_GATE_INP_SHEXP,
-                    gguf.MODEL_TENSOR.POS_EMBD,
-                    gguf.MODEL_TENSOR.TOKEN_TYPES,
-                    gguf.MODEL_TENSOR.SSM_CONV1D,
-                    gguf.MODEL_TENSOR.SHORTCONV_CONV,
-                    gguf.MODEL_TENSOR.TIME_MIX_FIRST,
-                    gguf.MODEL_TENSOR.TIME_MIX_W1,
-                    gguf.MODEL_TENSOR.TIME_MIX_W2,
-                    gguf.MODEL_TENSOR.TIME_MIX_DECAY_W1,
-                    gguf.MODEL_TENSOR.TIME_MIX_DECAY_W2,
-                    gguf.MODEL_TENSOR.TIME_MIX_LERP_FUSED,
-                    gguf.MODEL_TENSOR.POSNET_NORM1,
-                    gguf.MODEL_TENSOR.POSNET_NORM2,
-                    gguf.MODEL_TENSOR.V_ENC_EMBD_POS,
-                    gguf.MODEL_TENSOR.A_ENC_EMBD_POS,
-                    gguf.MODEL_TENSOR.ALTUP_CORRECT_COEF,
-                    gguf.MODEL_TENSOR.ALTUP_PREDICT_COEF,
-                    # Kimi KDA conv weights should be F32
-                    gguf.MODEL_TENSOR.SSM_CONV1D_Q,
-                    gguf.MODEL_TENSOR.SSM_CONV1D_K,
-                    gguf.MODEL_TENSOR.SSM_CONV1D_V,
-                    # DSA indexer weights should be F32
-                    gguf.MODEL_TENSOR.INDEXER_PROJ,
-                )
-            )
-            or new_name[-7:] not in (".weight", ".lora_a", ".lora_b")
-        )
-
    def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims: int) -> gguf.GGMLQuantizationType | bool:
-        if name in self._fp8_e4m3_preserved and n_dims >= 2:
-            if self._tensor_requires_f32(new_name, bid, n_dims):
-                return gguf.GGMLQuantizationType.F32
-            return gguf.GGMLQuantizationType.F8_E4M3
+        del new_name, bid  # unused
        # Force FP8-original tensors to Q8_0 when requested; Q8_0 is faster than F16/BF16.
        if self._fp8_as_q8 and name in self._fp8_dequantized and n_dims >= 2:
            return gguf.GGMLQuantizationType.Q8_0
@@ -1277,7 +1035,7 @@ class ModelBase:
        expert_scales: dict[tuple[int, str], list[tuple[int, float]]] = {}
        expert_input_scales: dict[tuple[int, str], list[tuple[int, float]]] = {}
        expert_shapes: dict[tuple[int, str], list[int]] = {}
-        n_experts = self.find_hparam(["num_local_experts", "num_experts", "n_routed_experts"], optional=True) or 0
+        n_experts = self.find_hparam(["num_local_experts", "num_experts"], optional=True) or 0
        consumed: list[str] = []

        for name in self.model_tensors.keys():
@@ -1288,17 +1046,14 @@ class ModelBase:
            input_scale_name = name.replace(".weight", ".input_scale")
            if scale_name not in self.model_tensors:
                continue
-            weight = self.model_tensors[name]()
-            scale = self.model_tensors[scale_name]()
+            # Force eager materialization of lazy tensors
+            weight = LazyTorchTensor.to_eager(self.model_tensors[name]())
+            scale = LazyTorchTensor.to_eager(self.model_tensors[scale_name]())

-            # Leave FP8 weights and their scales for _prepare_fp8_e4m3_tensors.
-            if weight.dtype in (torch.float8_e4m3fn, torch.float8_e5m2) or scale.ndim < 2:
+            # Skip non-NVFP4 tensors(e.g. 1D scale, or float8 weight)
+            if scale.ndim < 2 or weight.dtype in (torch.float8_e4m3fn, torch.float8_e5m2):
                continue
-            if weight.ndim != 2 or scale.shape[0] != weight.shape[0] or scale.shape[1] * 8 != weight.shape[1]:
-                raise ValueError(f"NVFP4 weight {name!r} has incompatible shapes: weight {list(weight.shape)}, scale {list(scale.shape)}")

-            weight = LazyTorchTensor.to_eager(weight)
-            scale = LazyTorchTensor.to_eager(scale)
            scale2 = LazyTorchTensor.to_eager(self.model_tensors.get(scale2_name, lambda: torch.tensor(1.0))())
            input_scale = LazyTorchTensor.to_eager(self.model_tensors.get(input_scale_name, lambda: torch.tensor(1.0))())

@@ -1454,10 +1209,8 @@ class ModelBase:
                            self.model_tensors[input_scale_name] = inverse_scale(self.model_tensors.pop(name))
            self._generate_nvfp4_tensors()

-        self._prepare_fp8_e4m3_tensors()
        self.dequant_model()

-        self.prepare_gate_up_fusion()
        self.prepare_qkv_fusion()

        # Handle empty tensor_map for models with block_count=0 (like MobileNetV5)
@@ -1471,21 +1224,10 @@ class ModelBase:
            if name.endswith((".attention.masked_bias", ".attention.bias", ".rotary_emb.inv_freq")):
                continue

-            # FP8 transforms must only reorder values; norms and vectors use numeric F32 transforms.
-            preserve_fp8_e4m3 = self._preserve_fp8 and data_torch.dtype == torch.float8_e4m3fn and data_torch.ndim >= 2 and not name.endswith(("norm.weight", "norm_f.weight"))
-            if preserve_fp8_e4m3:
-                for new_name in self._map_fp8_weight_names(name):
-                    mapped_bid = next((int(part) for part in new_name.split(".") if part.isdecimal()), None)
-                    if self._tensor_requires_f32(new_name, mapped_bid, data_torch.ndim):
-                        preserve_fp8_e4m3 = False
-                        break
-                if preserve_fp8_e4m3:
-                    self._fp8_e4m3_preserved.add(name)
-
            old_dtype = data_torch.dtype

            # convert any unsupported data types to float32
-            if data_torch.dtype not in (torch.float16, torch.float32) and name not in self._fp8_e4m3_preserved:
+            if data_torch.dtype not in (torch.float16, torch.float32):
                data_torch = data_torch.to(torch.float32)

            # use the first number-like part of the tensor name as the block id
@@ -1496,14 +1238,51 @@ class ModelBase:
                    break

            for new_name, data_torch in (self.modify_tensors(data_torch, name, bid)):
-                n_dims = len(data_torch.shape)
+                # TODO: why do we squeeze here?
+                # data = data_torch.squeeze().numpy()
+                data = data_torch.numpy()
+
+                n_dims = len(data.shape)
                data_qtype: gguf.GGMLQuantizationType | bool = self.tensor_force_quant(name, new_name, bid, n_dims)

                # Most of the codebase that takes in 1D tensors or norms only handles F32 tensors
                if n_dims <= 1 or new_name.endswith("_norm.weight"):
                    data_qtype = gguf.GGMLQuantizationType.F32

-                if data_qtype is False and self._tensor_requires_f32(new_name, bid, n_dims):
+                # Conditions should closely match those in llama_model_quantize_internal in llama.cpp
+                # Some tensor types are always in float32
+                if data_qtype is False and (
+                    any(
+                        self.match_model_tensor_name(new_name, key, bid)
+                        for key in (
+                            gguf.MODEL_TENSOR.FFN_GATE_INP,
+                            gguf.MODEL_TENSOR.FFN_GATE_INP_SHEXP,
+                            gguf.MODEL_TENSOR.POS_EMBD,
+                            gguf.MODEL_TENSOR.TOKEN_TYPES,
+                            gguf.MODEL_TENSOR.SSM_CONV1D,
+                            gguf.MODEL_TENSOR.SHORTCONV_CONV,
+                            gguf.MODEL_TENSOR.TIME_MIX_FIRST,
+                            gguf.MODEL_TENSOR.TIME_MIX_W1,
+                            gguf.MODEL_TENSOR.TIME_MIX_W2,
+                            gguf.MODEL_TENSOR.TIME_MIX_DECAY_W1,
+                            gguf.MODEL_TENSOR.TIME_MIX_DECAY_W2,
+                            gguf.MODEL_TENSOR.TIME_MIX_LERP_FUSED,
+                            gguf.MODEL_TENSOR.POSNET_NORM1,
+                            gguf.MODEL_TENSOR.POSNET_NORM2,
+                            gguf.MODEL_TENSOR.V_ENC_EMBD_POS,
+                            gguf.MODEL_TENSOR.A_ENC_EMBD_POS,
+                            gguf.MODEL_TENSOR.ALTUP_CORRECT_COEF,
+                            gguf.MODEL_TENSOR.ALTUP_PREDICT_COEF,
+                            # Kimi KDA conv weights should be F32
+                            gguf.MODEL_TENSOR.SSM_CONV1D_Q,
+                            gguf.MODEL_TENSOR.SSM_CONV1D_K,
+                            gguf.MODEL_TENSOR.SSM_CONV1D_V,
+                            # DSA indexer weights should be F32
+                            gguf.MODEL_TENSOR.INDEXER_PROJ,
+                        )
+                    )
+                    or new_name[-7:] not in (".weight", ".lora_a", ".lora_b")
+                ):
                    data_qtype = gguf.GGMLQuantizationType.F32

                if data_qtype is False and any(
@@ -1541,22 +1320,16 @@ class ModelBase:
                    else:
                        raise ValueError(f"Unknown file type: {self.ftype.name}")

-                if data_torch.dtype == torch.float8_e4m3fn:
-                    data = data_torch.view(torch.uint8).numpy() if data_qtype == gguf.GGMLQuantizationType.F8_E4M3 else data_torch.float().numpy()
-                else:
-                    data = data_torch.numpy()
-
                # a chunked tensor quantizes as one chunk at a time, while it is written
                quantize = data.quantize if isinstance(data, gguf.LazyChunkedTensor) else (
                    lambda qtype, d=data: gguf.quants.quantize(d, qtype))

-                if not (data_qtype == gguf.GGMLQuantizationType.F8_E4M3 and data.dtype == np.uint8):
-                    try:
-                        data = quantize(data_qtype)
-                    except gguf.QuantError as e:
-                        logger.warning("%s, %s", e, "falling back to F16")
-                        data_qtype = gguf.GGMLQuantizationType.F16
-                        data = quantize(data_qtype)
+                try:
+                    data = quantize(data_qtype)
+                except gguf.QuantError as e:
+                    logger.warning("%s, %s", e, "falling back to F16")
+                    data_qtype = gguf.GGMLQuantizationType.F16
+                    data = quantize(data_qtype)

                shape = gguf.quant_shape_from_byte_shape(data.shape, data_qtype) if data.dtype == np.uint8 else data.shape

@@ -1597,8 +1370,6 @@ class ModelBase:
                self.ftype = gguf.LlamaFileType.MOSTLY_NVFP4
            elif self._is_mxfp4:
                self.ftype = gguf.LlamaFileType.MOSTLY_MXFP4_MOE
-            elif self._fp8_e4m3_preserved:
-                self.ftype = gguf.LlamaFileType.MOSTLY_F8_E4M3

        # Generate parameter weight class (useful for leader boards) if not yet determined
        if self.metadata.size_label is None and total_params > 0:
@@ -1930,7 +1701,7 @@ class TextModel(ModelBase):
            self.gguf_writer.add_expert_group_used_count(n_group_used)
            logger.info(f"gguf: expert groups used count = {n_group_used}")

-        if (score_func := self.find_hparam(["score_function", "scoring_func", "score_func", "moe_router_activation", "moe_router_activation_func", "expert_selection_fn", "router_score_func", "router_score_function"], optional=True)) is not None:
+        if (score_func := self.find_hparam(["score_function", "scoring_func", "score_func", "moe_router_activation", "moe_router_activation_func", "expert_selection_fn", "router_score_func"], optional=True)) is not None:
            if score_func == "sigmoid":
                self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SIGMOID)
            elif score_func == "softmax":
@@ -2066,9 +1837,6 @@ class TextModel(ModelBase):
        if chkhsh == "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35":
            # ref: https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0
            res = "minerva-7b"
-        if chkhsh == "6e4d72c6e5260b90606e26a0931fa15e4b42ec6e97719d12919f2b1f93478262":
-            # ref: https://huggingface.co/domyn/Domyn-Small-v1.0
-            res = "domyn-small"
        if chkhsh == "7e57df22b1fe23a7b1e1c7f3dc4e3f96d43a4eb0836d0c6bdc3436d7b2f1c664":
            # ref: https://huggingface.co/tencent/Hunyuan-A13B-Instruct
            res = "hunyuan"
@@ -2171,9 +1939,6 @@ class TextModel(ModelBase):
        if chkhsh == "52df12b4c8d4176e7481aab4b6e8454d1fd0a210a04a574f6d4e067d10e23c3e":
            # ref: https://huggingface.co/CohereLabs/North-Mini-Code-1.0
            res = "cohere2moe"
-        if chkhsh == "d5420e601e979078c670bc503d9ac411c311fa76b76a76267c31e0d6ba816637":
-            # ref: https://huggingface.co/RecursiveSI/flint
-            res = "flint"
        if chkhsh == "e636dc30a262dcc0d8c323492e32ae2b70728f4df7dfe9737d9f920a282b8aea":
            # ref: https://huggingface.co/Qwen/Qwen1.5-7B
            res = "qwen2"
@@ -2901,13 +2666,7 @@ class TextModel(ModelBase):
        )

        if vocab.tokenizer_type == MistralTokenizerType.tekken:
-            with open(vocab.fname_tokenizer, encoding="utf-8") as f:
-                pattern = json.load(f)["config"]["pattern"]
-            # newer tekken.json split numbers into right-aligned groups of 3 digits instead of single digits
-            if "\\p{N}{1,3}(?=(?:\\p{N}{3})*(?!\\p{N}))" in pattern:
-                self.gguf_writer.add_tokenizer_pre("tekken-num3")
-            else:
-                self.gguf_writer.add_tokenizer_pre("tekken")
+            self.gguf_writer.add_tokenizer_pre("tekken")
            self.gguf_writer.add_token_merges(
                vocab.extract_vocab_merges_from_model()
            )
@@ -3384,7 +3143,7 @@ def get_model_architecture(hparams: dict[str, Any], model_type: ModelType) -> st
    # Kimi-K3's text_config reports "KimiLinearForCausalLM", which is the older
    # Kimi-Linear-48B architecture and cannot load K3 (no attention residuals,
    # latent MoE, situ, ...). Route on the top-level architecture instead.
-    if model_type == ModelType.TEXT and arch in ("NEOChatModel", "StepVLForConditionalGeneration", "Sarashina2VisionForCausalLM", "Exaone4_5_ForConditionalGeneration", "Step3p7ForConditionalGeneration", "KimiK3ForConditionalGeneration"):
+    if model_type == ModelType.TEXT and arch in ("StepVLForConditionalGeneration", "Sarashina2VisionForCausalLM", "Exaone4_5_ForConditionalGeneration", "Step3p7ForConditionalGeneration", "KimiK3ForConditionalGeneration"):
        return arch

    # if "architectures" is found in the sub-config, use that instead
@@ -236,12 +236,6 @@ class DeepseekModel(TextModel):
 class DeepseekV2Model(TextModel):
    model_arch = gguf.MODEL_ARCH.DEEPSEEK2

-    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
-        # MLA projections do not have runtime weight-scale support.
-        if name.endswith((".q_a_proj.weight", ".q_b_proj.weight", ".kv_a_proj_with_mqa.weight", ".kv_b_proj.weight")):
-            return ()
-        return super()._map_fp8_weight_names(name)
-
    # TODO @ngxson : remove this when we support MTP for deepseek models
    skip_mtp = True

@@ -809,6 +803,9 @@ class DeepseekV4Model(TextModel):
        self._dsv4_mxfp4_generated = True
        return ()

+    def _format_dsv4_tensor_name(self, key: gguf.MODEL_TENSOR, bid: int | None, suffix: str = ".weight") -> str:
+        return self.format_tensor_name(key, bid, suffix)
+
    def _map_dsv4_tensor_name(self, name: str, bid: int | None) -> tuple[gguf.MODEL_TENSOR, str]:
        root_map: dict[str, tuple[gguf.MODEL_TENSOR, str]] = {
            "embed.weight": (gguf.MODEL_TENSOR.TOKEN_EMBD, ".weight"),
@@ -882,17 +879,17 @@ class DeepseekV4Model(TextModel):

    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
        if re.match(r"layers\.\d+\.ffn\.experts\.\d+\.w[123]\.(weight|scale)$", name):
-            return
+            return []

        # hash layers route text tokens via tid2eid and image tokens via bias_vl; gate.bias is unused
        if name.endswith(".ffn.gate.bias") and bid is not None and bid < self.hparams["num_hash_layers"]:
-            return
+            return []

        tensor_key, suffix = self._map_dsv4_tensor_name(name, bid)
        if tensor_key == gguf.MODEL_TENSOR.FFN_GATE_TID2EID:
-            return
+            return []

-        yield self.format_tensor_name(tensor_key, bid, suffix), data_torch
+        return [(self._format_dsv4_tensor_name(tensor_key, bid, suffix), data_torch)]

    def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims: int) -> gguf.GGMLQuantizationType | bool:
        del bid  # unused
@@ -1,464 +0,0 @@
-from __future__ import annotations
-
-import json
-import math
-import re
-from pathlib import Path
-from typing import Any, Callable, Iterable, TYPE_CHECKING, cast
-
-import numpy as np
-import torch
-
-if TYPE_CHECKING:
-    from torch import Tensor
-
-from .base import LazyTorchTensor, ModelBase, TextModel, gguf
-from .deepseek import DeepseekV4FlashVisionModel, DeepseekV4Model
-
-
-# Engram hash constants from inference/engram.py, stored in the GGUF: the token map needs the HF tokenizers
-# normalizers and the multipliers numpy's PCG64 stream, neither of which the runtime has.
-
-def _is_prime(n: int) -> bool:
-    if n < 2 or n % 2 == 0:
-        return n == 2
-    return all(n % d for d in range(3, math.isqrt(n) + 1, 2))
-
-
-def engram_primes(n_layers: int, max_ngram_size: int, n_heads: int, vocab_size: int) -> list[list[list[int]]]:
-    """EngramLayout.from_args: [layer][ngram order 2..max][head] bucket moduli, never reused across layers."""
-    seen: set[int] = set()
-    out: list[list[list[int]]] = []
-    for _ in range(n_layers):
-        per_ngram = []
-        for _ in range(max_ngram_size - 1):
-            sizes, cur = [], vocab_size - 1
-            for _ in range(n_heads):
-                cand = cur + 1
-                while not _is_prime(cand) or cand in seen:
-                    cand += 1
-                cur = cand
-                seen.add(cur)
-                sizes.append(cur)
-            per_ngram.append(sizes)
-        out.append(per_ngram)
-    return out
-
-
-def engram_multipliers(layer_ids: list[int], max_ngram_size: int, compressed_vocab_size: int) -> list[list[int]]:
-    """compute_hash_multipliers: odd, bounded so that token_id * multiplier cannot overflow int64."""
-    bound = max(1, (np.iinfo(np.int64).max // compressed_vocab_size) // 2)
-    rows = []
-    for layer_id in layer_ids:
-        rng = np.random.default_rng(10007 * layer_id)
-        values = rng.integers(low=0, high=bound, size=(max_ngram_size,), dtype=np.int64)
-        rows.append([int(v) * 2 + 1 for v in values])
-    return rows
-
-
-def engram_compressed_token_map(tokenizer_dir: Path) -> tuple[list[int], int]:
-    """build_compressed_token_map: token id -> compressed id (tokens that normalize alike collapse)."""
-    from tokenizers import Regex, normalizers
-    from transformers import AutoTokenizer
-
-    tokenizer = AutoTokenizer.from_pretrained(tokenizer_dir)
-    sentinel = "\ue000"
-    normalizer = normalizers.Sequence([
-        normalizers.NFKC(),
-        normalizers.NFD(),
-        normalizers.StripAccents(),
-        normalizers.Lowercase(),
-        normalizers.Replace(Regex(r"[ \t\r\n]+"), " "),
-        normalizers.Replace(Regex(r"^ $"), sentinel),
-        normalizers.Strip(),
-        normalizers.Replace(sentinel, " "),
-    ])
-    backend = tokenizer.backend_tokenizer
-    key_to_new: dict[str, int] = {}
-    lookup = [0] * len(tokenizer)
-    for token_id in range(len(tokenizer)):
-        text = backend.decode([token_id], skip_special_tokens=False)
-        if "\ufffd" in text:
-            key = backend.id_to_token(token_id)
-        else:
-            normalized = normalizer.normalize_str(text)
-            key = normalized if normalized else text
-        lookup[token_id] = key_to_new.setdefault(key, len(key_to_new))
-    return lookup, len(key_to_new)
-
-
-class _StreamedBytes:
-    """Raw tensor bytes produced chunk by chunk at write time (an Engram table is ~200 GB in BF16)."""
-    def __init__(self, shape: tuple[int, ...], chunks: Callable[[], Iterable[np.ndarray]]):
-        self.shape = shape
-        self.dtype = np.dtype(np.uint8)
-        self.nbytes = int(np.prod(shape))
-        self._chunks = chunks
-
-    def tofile(self, fout) -> None:
-        written = 0
-        for chunk in self._chunks():
-            chunk = np.ascontiguousarray(chunk, dtype=np.uint8)
-            chunk.tofile(fout)
-            written += chunk.nbytes
-        if written != self.nbytes:
-            raise ValueError(f"streamed {written} bytes, expected {self.nbytes}")
-
-    def byteswap(self, inplace: bool = False):
-        raise NotImplementedError("streamed tensors are written little-endian only")
-
-
-def _e8m0_check(scale_u8: np.ndarray, name: str) -> None:
-    # e4m3 (|x| in [2^-9, 448]) * 2^(e - 127) is exact in BF16 for e in [3, 246]
-    if scale_u8.size and (int(scale_u8.min()) < 3 or int(scale_u8.max()) > 246):
-        raise ValueError(f"{name}: E8M0 scale exponents outside [3, 246], the dequantized values would not be exact")
-
-
-@ModelBase.register("DeepseekV41ForCausalLM")
-class DeepseekV41Model(DeepseekV4Model):
-    """DeepSeek-V4.1 text model, lossless: FP8 (e4m3, 32x32 ue8m0 blocks) linears dequantized to BF16 (F32 with
-    --outtype f32), routed experts copied bit-exact as MXFP4, Engram tables (e4m3 rows, one e8m0 scale per 32) as the
-    exact BF16 lookup (F32 with --outtype f32), router and Engram gate F32. DSpark (mtp.*) is not converted here
-    (--dspark, DeepseekV41DSparkModel); the vision tower goes to an mmproj (--mmproj, DeepseekV41VisionModel).
-    The Engram tables of all layers are written as one per_layer_token_embd (rows concatenated in layer order) with
-    the PLE keys of qwen4exp: absolute head offsets and bucket counts, so the runtime can read it lazily."""
-    model_arch = gguf.MODEL_ARCH.DEEPSEEK41
-    supports_mtp_export = False
-
-    _ENGRAM_EMBED_RE = re.compile(r"layers\.(\d+)\.engram\.embed\.(weight|scale)$")
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.hparams.setdefault("num_hash_layers", 0)  # no hash-routed layers in V4.1
-        self._f32 = self.ftype == gguf.LlamaFileType.ALL_F32
-        qc = self.hparams.get("quantization_config") or {}
-        self._fp8_block = tuple(qc.get("weight_block_size") or (32, 32))
-        if qc.get("scale_fmt", "ue8m0") != "ue8m0":
-            raise ValueError(f"unsupported FP8 scale format {qc.get('scale_fmt')!r}")
-        if self.hparams.get("gate_temp", 1.0) != 1.0:
-            raise ValueError("gate_temp != 1 is not supported (the router logits would need dividing by it)")
-
-        # V4.1 does not use the DeepSeek-V4 template; a template shipped with the model is still added by set_vocab
-        self.gguf_writer.remove_key(gguf.Keys.Tokenizer.CHAT_TEMPLATE)
-
-    def index_tensors(self, remote_hf_model_id: str | None = None) -> dict[str, Callable[[], Tensor]]:
-        # num_hidden_layers lives in text_config, which is merged only after indexing; MTP stages are skipped anyway
-        return TextModel.index_tensors(self, remote_hf_model_id=remote_hf_model_id)
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        hp = self.hparams
-        w = self.gguf_writer
-        w.add_attention_kv_source_layers(hp["kv_source_layer_ids"])
-        w.add_indexer_source_layers(hp["index_source_layer_ids"])
-        if hp.get("candidate_source_layer_id", -1) >= 0:
-            w.add_indexer_candidate_source_layer(hp["candidate_source_layer_id"])
-            w.add_indexer_candidate_top_k_blocks(hp["candidate_topk_blocks"])
-            w.add_indexer_candidate_block_size(hp["candidate_block_size"])
-
-        # GGUF cannot hold empty arrays: the Engram keys are omitted when the model has no Engram layers
-        layer_ids = list(hp.get("engram_layer_ids") or [])
-        if layer_ids:
-            max_ngram, n_heads = hp["engram_max_ngram_size"], hp["engram_n_heads"]
-            primes = engram_primes(len(layer_ids), max_ngram, n_heads, hp["engram_vocab_size"])
-            for li, layer_primes in enumerate(primes):
-                if sum(p for per in layer_primes for p in per) != hp["engram_num_embeddings"][li]:
-                    raise ValueError(f"engram layer {layer_ids[li]}: prime sum != num_embeddings")
-            token_map, cvocab = engram_compressed_token_map(self.dir_model)
-            if cvocab != hp["engram_compressed_vocab_size"] or len(token_map) != hp["vocab_size"]:
-                raise ValueError(f"compressed token map ({len(token_map)} -> {cvocab}) does not match the config")
-            # absolute row offsets into the concatenated table
-            sizes = [p for layer in primes for per in layer for p in per]
-            offsets = [sum(sizes[:i]) for i in range(len(sizes))]
-            w.add_key_value(gguf.Keys.PerLayerEmbedding.LAYERS.format(arch=w.arch), layer_ids,
-                            gguf.GGUFValueType.ARRAY, gguf.GGUFValueType.UINT32)
-            w.add_ple_ngram_size(max_ngram)
-            w.add_ple_heads_per_ngram(n_heads)
-            w.add_embedding_length_per_layer_input(hp["engram_head_dim"])
-            w.add_ple_layer_multipliers([m for row in engram_multipliers(layer_ids, max_ngram, cvocab) for m in row])
-            w.add_ple_head_offsets(offsets)
-            w.add_ple_head_vocab_sizes(sizes)
-            w.add_engram_pad_token_id(hp["engram_pad_token_id"])
-            w.add_engram_token_map(token_map)
-
-    def dequant_model(self):
-        bo, bi = self._fp8_block
-
-        def dequant(weight: Tensor, scale: Tensor) -> Tensor:
-            out_f, in_f = weight.shape
-            s = self._e8m0_to_float(scale).repeat_interleave(bo, 0)[:out_f].repeat_interleave(bi, 1)[:, :in_f]
-            return weight.float() * s
-
-        for name in [n for n in self.model_tensors if n.endswith(".scale") and not self._ENGRAM_EMBED_RE.match(n)]:
-            weight_name = name.removesuffix(".scale") + ".weight"
-            if weight_name not in self.model_tensors or self.model_tensors[weight_name]().dtype not in self._float8_dtypes():
-                continue
-            w, s = self.model_tensors[weight_name], self.model_tensors.pop(name)
-            out_f, in_f = w().shape
-            if tuple(s().shape) != (-(-out_f // bo), -(-in_f // bi)):
-                raise ValueError(f"{weight_name}: scale shape {tuple(s().shape)} does not match {(bo, bi)} blocks")
-            _e8m0_check(LazyTorchTensor.to_eager(s()).view(torch.uint8).numpy(), name)
-            self.model_tensors[weight_name] = lambda w=w, s=s: dequant(w(), s())
-            self._dsv4_fp8_dequantized.add(weight_name)
-
-    def _tensor_part(self, name: str) -> Path:
-        index = self.dir_model / "model.safetensors.index.json"
-        if not index.is_file():
-            return self.dir_model / "model.safetensors"
-        with open(index, "r", encoding="utf-8") as f:
-            return self.dir_model / json.load(f)["weight_map"][name]
-
-    def _write_mxfp4_experts(self, bid: int, proj: str, key: gguf.MODEL_TENSOR, n_expert: int | None = None) -> None:
-        n_expert = self.hparams["n_routed_experts"] if n_expert is None else n_expert
-        names = [f"layers.{bid}.ffn.experts.{e}.{proj}" for e in range(n_expert)]
-        pairs = [(self.model_tensors.pop(n + ".weight"), self.model_tensors.pop(n + ".scale")) for n in names]
-        rows, packed_cols = pairs[0][0]().shape
-
-        def chunks() -> Iterable[np.ndarray]:
-            for w, s in pairs:
-                yield self.repack_mxfp4_blocks(LazyTorchTensor.to_eager(w()), LazyTorchTensor.to_eager(s()))
-        data = cast(np.ndarray, _StreamedBytes((len(pairs), rows, packed_cols * 2 // 32 * 17), chunks))
-        self.gguf_writer.add_tensor(self.format_tensor_name(key, bid), data, raw_dtype=gguf.GGMLQuantizationType.MXFP4)
-
-    def _write_engram_tables(self, layer_ids: list[int]) -> None:
-        # official lookup e4m3 * 2^(e8m0 - 127), exact in BF16
-        parts, dim = [], None
-        for bid in layer_ids:
-            wn, sn = f"layers.{bid}.engram.embed.weight", f"layers.{bid}.engram.embed.scale"
-            rows, d = self.model_tensors.pop(wn)().shape
-            if d % 32 != 0 or tuple(self.model_tensors.pop(sn)().shape) != (rows, d // 32) or d != (dim or d):
-                raise ValueError(f"{wn}: expected one e8m0 scale per 32 elements and the same width in every layer")
-            if rows != self.hparams["engram_num_embeddings"][len(parts)]:
-                raise ValueError(f"{wn}: {rows} rows, config says {self.hparams['engram_num_embeddings'][len(parts)]}")
-            dim = d
-            parts.append((wn, sn, rows, self._tensor_part(wn), self._tensor_part(sn)))
-        assert dim is not None
-        name = self.format_tensor_name(gguf.MODEL_TENSOR.PER_LAYER_TOKEN_EMBD)
-        n_rows = sum(p[2] for p in parts)
-
-        def chunks() -> Iterable[np.ndarray]:
-            from safetensors import safe_open
-            for wn, sn, rows, wpart, spart in parts:
-                with safe_open(wpart, framework="pt") as fw, safe_open(spart, framework="pt") as fs:
-                    ws, ss = fw.get_slice(wn), fs.get_slice(sn)
-                    for r0 in range(0, rows, 1 << 18):
-                        q = ws[r0:r0 + (1 << 18)].view(torch.uint8).reshape(-1, 32)
-                        e = ss[r0:r0 + (1 << 18)].view(torch.uint8).reshape(-1, 1)
-                        _e8m0_check(e.numpy(), sn)
-                        v = (q.view(torch.float8_e4m3fn).float() * torch.exp2(e.float() - 127)).reshape(-1, dim)
-                        yield (v if self._f32 else v.bfloat16()).view(torch.uint8).numpy()
-        qt = gguf.GGMLQuantizationType.F32 if self._f32 else gguf.GGMLQuantizationType.BF16
-        data = cast(np.ndarray, _StreamedBytes((n_rows, dim * (4 if self._f32 else 2)), chunks))
-        self.gguf_writer.add_tensor(name, data, raw_dtype=qt)
-
-    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
-        if not self._dsv4_mxfp4_generated:
-            for bid in range(self.block_count):
-                self._write_mxfp4_experts(bid, "w1", gguf.MODEL_TENSOR.FFN_GATE_EXP)
-                self._write_mxfp4_experts(bid, "w2", gguf.MODEL_TENSOR.FFN_DOWN_EXP)
-                self._write_mxfp4_experts(bid, "w3", gguf.MODEL_TENSOR.FFN_UP_EXP)
-            if self.hparams.get("engram_layer_ids"):
-                self._write_engram_tables(list(self.hparams["engram_layer_ids"]))
-            self._dsv4_mxfp4_generated = True
-        return ()
-
-    _LAYER_MAP: dict[str, tuple[gguf.MODEL_TENSOR, str]] = {
-        "attn.indexer.wk.weight": (gguf.MODEL_TENSOR.INDEXER_ATTN_K, ".weight"),
-        "attn.indexer.k_norm.weight": (gguf.MODEL_TENSOR.INDEXER_K_NORM, ".weight"),
-        "engram.wkv.weight": (gguf.MODEL_TENSOR.ENGRAM_KV, ".weight"),
-    }
-
-    def _map_dsv4_tensor_name(self, name: str, bid: int | None) -> tuple[gguf.MODEL_TENSOR, str]:
-        m = re.match(r"layers\.\d+\.(.+)$", name)
-        if m is not None and m.group(1) in self._LAYER_MAP:
-            return self._LAYER_MAP[m.group(1)]
-        return super()._map_dsv4_tensor_name(name, bid)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.endswith(".engram.k_weight"):
-            return  # folded into engram_gate
-        if name.endswith(".engram.q_weight"):
-            # Engram gate weight = q_weight * k_weight (bf16 * bf16 is exact in f32)
-            k = self.model_tensors[name.removesuffix("q_weight") + "k_weight"]
-            yield self.format_tensor_name(gguf.MODEL_TENSOR.ENGRAM_GATE, bid), data_torch.float() * k().float()
-            return
-        yield from super().modify_tensors(data_torch, name, bid)
-
-    def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims: int) -> gguf.GGMLQuantizationType | bool:
-        if new_name.endswith((".engram_gate.weight", ".ffn_gate_inp.weight")) or name in self._dsv4_f32_tensors:
-            return gguf.GGMLQuantizationType.F32
-        if n_dims >= 2 and (name in self._dsv4_fp8_dequantized or name in self._dsv4_bf16_tensors):
-            return gguf.GGMLQuantizationType.F32 if self._f32 else gguf.GGMLQuantizationType.BF16
-        return False
-
-
-class DeepseekV41DSparkModel(DeepseekV41Model):
-    """DeepSeek-V4.1 DSpark speculative draft (mtp.0..2 of the checkpoint) as a DFLASH GGUF (`--dspark`).
-
-    Layout = the V4 DSpark draft (conversion/deepseek.py DeepseekV4DSparkModel) with V4.1's differences flagged by
-    `dflash.hyper_connection.single_pass`: stages blk.0..2 (MXFP4 experts bit-exact, dense BF16), mtp.0.main_proj /
-    main_norm -> fc / enc.output_norm, mtp.2.norm -> output_norm, markov_head.embed / head -> markov_w1 / markov_w2,
-    confidence_head.proj -> conf_proj (F32), no hc_head. Token embeddings and the lm head are the target's (shared at
-    runtime through ctx_other). target_layers are the V4.1 ids as-is: the draft reads the attention input of those
-    layers, which is what llama_set_embeddings_layer_inp extracts.
-    """
-    model_arch = gguf.MODEL_ARCH.DFLASH
-    mtp_only = True
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        hp = self.hparams
-        self._n_main = hp["num_hidden_layers"]
-        self._n_stage = hp["num_nextn_predict_layers"]
-        if self._n_stage <= 0 or "dspark_block_size" not in hp:
-            raise ValueError("DeepSeek-V4.1 DSpark export needs num_nextn_predict_layers > 0 and the dspark_* config")
-        # stages keep main-model block ids (mtp.M -> layers.n_main+M); format_tensor_name() maps them back
-        self.tensor_map = gguf.get_tensor_name_map(self.model_arch, self._n_stage)
-
-    def index_tensors(self, remote_hf_model_id: str | None = None) -> dict[str, Callable[[], Tensor]]:
-        hp = {**self.hparams, **self.hparams.get("text_config", {})}  # text_config is merged only after indexing
-        type(self)._dsv4_main_layers = hp["num_hidden_layers"]
-        type(self)._dsv4_nextn_layers = hp.get("num_nextn_predict_layers", 0)
-        return TextModel.index_tensors(self, remote_hf_model_id=remote_hf_model_id)
-
-    @classmethod
-    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
-        name, gen = item
-        m = re.match(r"mtp\.(\d+)\.(.+)$", name)
-        if m is None:
-            return None  # shared with the target at runtime
-        stage, rest = int(m.group(1)), m.group(2)
-        assert cls._dsv4_main_layers is not None
-        if stage >= cls._dsv4_nextn_layers:
-            raise ValueError(f"Unexpected DSpark stage {stage}")
-        if rest.startswith(("markov_head.", "confidence_head.")):
-            return f"dspark.{rest}", gen
-        return f"layers.{cls._dsv4_main_layers + stage}.{rest}", gen
-
-    def format_tensor_name(self, key: gguf.MODEL_TENSOR, bid: int | None = None, suffix: str = ".weight") -> str:
-        if bid is not None and bid >= self._n_main:
-            bid -= self._n_main
-        return super().format_tensor_name(key, bid, suffix)
-
-    def set_vocab(self):
-        super().set_vocab()
-        self.gguf_writer.add_mask_token_id(self.hparams["dspark_noise_token_id"])
-
-    def set_gguf_parameters(self):
-        hp = self.hparams
-        n = self._n_stage
-        w = self.gguf_writer
-        w.add_block_count(n)
-        w.add_context_length(hp["max_position_embeddings"])
-        w.add_embedding_length(hp["hidden_size"])
-        w.add_feed_forward_length(hp["moe_intermediate_size"])
-        w.add_head_count(hp["num_attention_heads"])
-        w.add_head_count_kv(1)
-        w.add_key_length(hp["head_dim"])
-        w.add_value_length(hp["head_dim"])
-        w.add_rope_dimension_count(hp["qk_rope_head_dim"])
-        # compress-ratio-0 layers: plain rope at rope_theta, no YaRN
-        w.add_rope_freq_base(float(hp["rope_theta"]))
-        w.add_layer_norm_rms_eps(hp["rms_norm_eps"])
-        w.add_q_lora_rank(hp["q_lora_rank"])
-        w.add_sliding_window(hp["sliding_window"])
-        w.add_expert_count(hp["dspark_n_routed_experts"])
-        w.add_expert_used_count(hp["dspark_num_experts_per_tok"])
-        w.add_expert_feed_forward_length(hp["moe_intermediate_size"])
-        w.add_expert_shared_count(hp["n_shared_experts"])
-        w.add_expert_weights_scale(hp["routed_scaling_factor"])
-        w.add_expert_weights_norm(hp["norm_topk_prob"])
-        w.add_expert_gating_func(gguf.ExpertGatingFuncType.SQRTSOFTPLUS)
-        w.add_swiglu_clamp_exp([hp["swiglu_limit"]] * n)
-        w.add_swiglu_clamp_shexp([hp["swiglu_limit"]] * n)
-        w.add_attention_output_group_count(hp["o_groups"])
-        w.add_attention_output_lora_rank(hp["o_lora_rank"])
-        w.add_attention_compress_ratios([0] * n)
-        w.add_hyper_connection_count(hp["hc_mult"])
-        w.add_hyper_connection_sinkhorn_iterations(hp["hc_sinkhorn_iters"])
-        w.add_hyper_connection_epsilon(hp["hc_eps"])
-        w.add_hyper_connection_single_pass(True)
-        w.add_block_size(hp["dspark_block_size"])
-        w.add_target_layers(hp["dspark_target_layer_ids"])
-        w.add_sample_from_anchor(True)
-        w.add_has_confidence_head(True)
-        w.add_file_type(self.ftype)
-
-    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
-        # no Engram tables in the draft
-        if not self._dsv4_mxfp4_generated:
-            n_expert = self.hparams["dspark_n_routed_experts"]
-            for bid in range(self._n_main, self._n_main + self._n_stage):
-                self._write_mxfp4_experts(bid, "w1", gguf.MODEL_TENSOR.FFN_GATE_EXP, n_expert)
-                self._write_mxfp4_experts(bid, "w2", gguf.MODEL_TENSOR.FFN_DOWN_EXP, n_expert)
-                self._write_mxfp4_experts(bid, "w3", gguf.MODEL_TENSOR.FFN_UP_EXP, n_expert)
-            self._dsv4_mxfp4_generated = True
-        return ()
-
-    _DSPARK_ROOT: dict[str, gguf.MODEL_TENSOR] = {
-        "main_proj.weight": gguf.MODEL_TENSOR.FC,
-        "main_norm.weight": gguf.MODEL_TENSOR.ENC_OUTPUT_NORM,
-        "norm.weight": gguf.MODEL_TENSOR.OUTPUT_NORM,
-        "dspark.markov_head.embed.weight": gguf.MODEL_TENSOR.DSPARK_MARKOV_W1,
-        "dspark.markov_head.head.weight": gguf.MODEL_TENSOR.DSPARK_MARKOV_W2,
-        "dspark.confidence_head.proj.weight": gguf.MODEL_TENSOR.DSPARK_CONF_PROJ,
-    }
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.endswith(".ffn.gate.bias_vl"):
-            return
-        if name in self._DSPARK_ROOT:
-            yield self.format_tensor_name(self._DSPARK_ROOT[name], None), data_torch
-            return
-        m = re.match(r"layers\.\d+\.(main_proj\.weight|main_norm\.weight|norm\.weight)$", name)
-        if m is not None:
-            yield self.format_tensor_name(self._DSPARK_ROOT[m.group(1)], None), data_torch
-            return
-        yield from super().modify_tensors(data_torch, name, bid)
-
-    def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims: int) -> gguf.GGMLQuantizationType | bool:
-        if new_name == "conf_proj.weight":
-            return gguf.GGMLQuantizationType.F32
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
-
-
-@ModelBase.register("DeepseekV41ForCausalLM")
-@ModelBase.example("deepseek-ai/DeepSeek-V4.1-Flash-1017")
-class DeepseekV41VisionModel(DeepseekV4FlashVisionModel):
-    """DeepSeek-V4.1 vision tower (projector deepseek41v): the V4-Flash-Vision ViT + 3x3 aligner, with the V4.1 token
-    block ([START] ([IMAGE]*w [NEWLINE])*h [END], no pads) and resize plan. ref: inference/vision.py,
-    inference/image_processor.py of deepseek-ai/DeepSeek-V4.1-Flash-1017. The checkpoint config is HF-style
-    (vision_config); image_start / image_end / image_newline are written like V4's sentinels, there is no image_pad."""
-
-    def get_vision_config(self) -> dict[str, Any] | None:
-        vc = self.global_config.get("vision_config")
-        if not vc or vc.get("num_hidden_layers", 0) == 0:
-            raise ValueError("DeepseekV41VisionModel requires vision_config.num_hidden_layers > 0")
-        return {
-            "num_hidden_layers":   vc["num_hidden_layers"],
-            "hidden_size":         vc["hidden_size"],
-            "num_attention_heads": vc["num_attention_heads"],
-            "intermediate_size":   vc["intermediate_size"],
-            "patch_size":          vc["patch_size"],
-            # dynamic resolution; only used for compat / warmup
-            "image_size":          vc["patch_size"] * vc["downsample_ratio"] * 16,
-            "rope_theta":          vc.get("rope_theta", 10000.0),
-            "downsample_ratio":    vc["downsample_ratio"],
-            "min_pixels":          vc["min_pixels"],
-            "max_image_tokens":    vc["max_image_tokens"],
-            "max_wh_ratio":        vc.get("max_wh_ratio"),
-        }
-
-    def set_gguf_parameters(self):
-        assert self.hparams_vision is not None
-        super(DeepseekV4FlashVisionModel, self).set_gguf_parameters()
-        self.gguf_writer.add_clip_projector_type(gguf.VisionProjectorType.DEEPSEEK41V)
-        # pytorch default eps (vision.py), not the LLM rms_norm_eps (1e-20)
-        self.gguf_writer.add_vision_attention_layernorm_eps(1e-6)
-        self.gguf_writer.add_vision_use_silu(True)
-        self.gguf_writer.add_vision_projector_scale_factor(self.hparams_vision["downsample_ratio"])
-        self.gguf_writer.add_vision_min_pixels(self.hparams_vision["min_pixels"])
-        # hardcoded in clip.cpp (PROJECTOR_TYPE_DEEPSEEK41V); add GGUF keys if a later model differs
-        assert self.hparams_vision["max_image_tokens"] == 1024, self.hparams_vision["max_image_tokens"]
-        assert self.hparams_vision["max_wh_ratio"] is None, self.hparams_vision["max_wh_ratio"]
-        assert self.hparams_vision["rope_theta"] == 10000, self.hparams_vision["rope_theta"]
@@ -1,38 +0,0 @@
-from __future__ import annotations
-
-from .base import ModelBase, gguf
-from .qwen import Qwen2MoeModel
-
-
-@ModelBase.register("FlintForCausalLM")
-@ModelBase.example("RecursiveSI/flint")
-class FlintModel(Qwen2MoeModel):
-    model_arch = gguf.MODEL_ARCH.FLINT
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        dense = self.hparams["mlp_only_layers"]
-        if dense != list(range(len(dense))):
-            raise ValueError("Flint dense layers must precede the MoE layers")
-        self.gguf_writer.add_leading_dense_block_count(len(dense))
-        self.gguf_writer.add_expert_shared_count(
-            self.hparams["shared_expert_intermediate_size"] // self.hparams["moe_intermediate_size"]
-        )
-        self.gguf_writer.add_expert_weights_norm(self.hparams["norm_topk_prob"])
-        self.gguf_writer.add_expert_weights_scale(1.0)
-        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
-        self.gguf_writer.add_sliding_window_pattern([
-            kind == "sliding_attention" for kind in self.hparams["layer_types"]
-        ])
-        self.gguf_writer.add_rope_pattern([
-            i not in self.hparams["no_rope_layer_indices"] for i in range(self.block_count)
-        ])
-
-    def modify_tensors(self, data_torch, name, bid):
-        if name.endswith(".softmax_offset"):
-            name += ".weight"
-        if name.endswith(".gate.expert_bias"):
-            if not self.hparams["router_expert_bias"]:
-                return
-            name += ".bias"
-        yield from super().modify_tensors(data_torch, name, bid)
@@ -15,30 +15,25 @@ from .base import MmprojModel, ModelBase, TextModel, gguf, jinja_str_or_json, lo
 from .gemma import ConformerAudioModel


-def lfm2_feed_forward_length(hparams: dict[str, Any]) -> int:
-    ff_dim = hparams.get("block_ff_dim", hparams.get("intermediate_size"))
-    assert ff_dim is not None
-    auto_adjust_ff_dim = hparams["block_auto_adjust_ff_dim"]
-    ffn_dim_multiplier = hparams["block_ffn_dim_multiplier"]
-    multiple_of = hparams["block_multiple_of"]
-
-    if auto_adjust_ff_dim:
-        ff_dim = int(2 * ff_dim / 3)
-        # custom dim factor multiplier
-        if ffn_dim_multiplier is not None:
-            ff_dim = int(ffn_dim_multiplier * ff_dim)
-        ff_dim = multiple_of * ((ff_dim + multiple_of - 1) // multiple_of)
-
-    return ff_dim
-
-
@ModelBase.register("Lfm2ForCausalLM", "LFM2ForCausalLM")
@ModelBase.example("LiquidAI/LFM2-1.2B", "LiquidAI/LFM2.5-350M")
 class LFM2Model(TextModel):
    model_arch = gguf.MODEL_ARCH.LFM2

    def _add_feed_forward_length(self):
-        self.gguf_writer.add_feed_forward_length(lfm2_feed_forward_length(self.hparams))
+        ff_dim = self.find_hparam(["block_ff_dim", "intermediate_size"])
+        auto_adjust_ff_dim = self.hparams["block_auto_adjust_ff_dim"]
+        ffn_dim_multiplier = self.hparams["block_ffn_dim_multiplier"]
+        multiple_of = self.hparams["block_multiple_of"]
+
+        if auto_adjust_ff_dim:
+            ff_dim = int(2 * ff_dim / 3)
+            # custom dim factor multiplier
+            if ffn_dim_multiplier is not None:
+                ff_dim = int(ffn_dim_multiplier * ff_dim)
+            ff_dim = multiple_of * ((ff_dim + multiple_of - 1) // multiple_of)
+
+        self.gguf_writer.add_feed_forward_length(ff_dim)

    def set_gguf_parameters(self):
        # set num_key_value_heads only for attention layers
@@ -77,9 +72,7 @@ def _is_d1_checkpoint(dir_model: Path) -> bool:
    if not (dir_model / "config.json").is_file():
        return False
    with open(dir_model / "config.json", encoding="utf-8") as f:
-        config = json.load(f)
-    # d1 on LFM3 is converted by conversion/lfm3.py
-    return config.get("auto_map", {}).get("AutoModel", "").endswith(".D1Model") and not config.get("model_type", "").startswith("lfm3")
+        return json.load(f).get("auto_map", {}).get("AutoModel", "").endswith(".D1Model")


@ModelBase.register_hparams_loader(_is_d1_checkpoint)
@@ -1,629 +0,0 @@
-from __future__ import annotations
-
-import json
-import math
-
-from pathlib import Path
-from typing import Any, Callable, Iterable, TYPE_CHECKING
-
-import torch
-
-if TYPE_CHECKING:
-    from torch import Tensor
-
-from .base import MmprojModel, ModelBase, TextModel, gguf, logger
-from .lfm2 import D1Model, lfm2_feed_forward_length
-
-
-@ModelBase.register("Lfm3ForCausalLM", "Lfm3VlForConditionalGeneration", "Lfm3OmniForConditionalGeneration")
-@ModelBase.example("LiquidAI/LFM3-1.7B", "LiquidAI/LFM3-1.7B-Base")
-class LFM3Model(TextModel):
-    model_arch = gguf.MODEL_ARCH.LFM3
-
-    # LFM3-Audio/Omni: encoder, projector and audio-output heads live next to the text model
-    _media_prefixes = ("encoder.", "mlp.", "codebook_embedding.", "depth_embeddings.", "depth_linear.", "depthformer.",
-                       "mask_embedding", "model.codebook_embedding.", "model.depth_decoder.")
-
-    @classmethod
-    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
-        name, gen = item
-
-        if name.startswith(cls._media_prefixes):
-            return None
-
-        name = name.removeprefix("lfm.")
-
-        return super().filter_tensors((name, gen))
-
-    # LFM3-Audio live ASR prompt of the model card, the audio follows in the assistant turn
-    _asr_template = (
-        "{{- bos_token -}}"
-        "{{- '<|im_start|>system\\nPerform live ASR.<|im_end|>\\n' -}}"
-        "{%- if add_generation_prompt -%}{{- '<|im_start|>assistant\\n' -}}{%- endif -%}"
-    )
-
-    # TTS prompt of the model card: a language code is looked up, anything else is taken as the language name
-    _tts_langs = {
-        "ar": "Arabic", "da": "Danish", "de": "German", "el": "Greek", "en": "American English", "en-us": "American English",
-        "en-gb": "British English", "fi": "Finnish", "fr": "French", "he": "Hebrew", "it": "Italian", "ja": "Japanese",
-        "ko": "Korean", "nl": "Dutch", "pl": "Polish", "pt": "Portuguese", "es": "Spanish", "sv": "Swedish", "tr": "Turkish",
-        "zh": "Chinese",
-    }
-
-    # "language" and "voice" are optional, the text to speak is the first message
-    def _tts_template(self, think: bool) -> str:
-        return (
-            "{%- set names = " + json.dumps(self._tts_langs) + " -%}"
-            "{%- set code = (language | lower) if language is defined and language else 'en' -%}"
-            "{%- set lang = names[code] if code in names else language -%}"
-            "{%- set v = (voice | lower) if voice is defined and voice and (voice | lower) in ['female', 'male'] else 'female' -%}"
-            "{{- bos_token -}}"
-            "{{- '<|im_start|>system\\nPerform TTS in ' + lang + ' using a ' + v + ' voice.<|im_end|>\\n' -}}"
-            "{{- '<|im_start|>user\\n' + messages[0]['content'] + '<|im_end|>\\n' -}}"
-            "{{- '<|im_start|>assistant\\n" + ("<think></think>" if think else "") + "<|audio_start|>' -}}"
-        )
-
-    def set_vocab(self):
-        super().set_vocab()
-        # hparams hold the merged text config, the checkpoint's own model_type is in config.json
-        with open(self.dir_model / "config.json", encoding="utf-8") as f:
-            model_type = json.load(f).get("model_type")
-        is_audio = model_type == "lfm3_audio"
-        is_omni  = model_type == "lfm3_omni"
-        templates = []
-        if is_audio:
-            templates.append({"name": "asr", "template": self._asr_template})
-        if is_audio or is_omni:
-            # LFM3-Omni opens every assistant turn with an empty think block
-            templates.append({"name": "tts", "template": self._tts_template(think=is_omni)})
-        if templates:
-            self.gguf_writer.add_chat_template(templates)
-
-    def set_gguf_parameters(self):
-        layer_types = self.hparams["layer_types"]
-        n_kv_heads = self.hparams["num_key_value_heads"]
-
-        # set num_key_value_heads only for attention layers
-        self.hparams["num_key_value_heads"] = [
-            n_kv_heads if lt != "conv" else 0
-            for lt in layer_types
-        ]
-
-        super().set_gguf_parameters()
-
-        self.gguf_writer.add_rope_dimension_count(self.hparams["head_dim"])
-
-        self.gguf_writer.add_sliding_window_pattern([lt == "sliding_attention" for lt in layer_types])
-        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
-
-        self.gguf_writer.add_shortconv_l_cache(self.hparams["conv_L_cache"])
-
-    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
-        # Partial RoPE is represented by masking trailing frequency pairs.
-        # rope_parameters is keyed by layer type; only full_attention layers use partial RoPE.
-        rope_parameters = self.hparams["rope_parameters"]["full_attention"]
-        partial = rope_parameters.get("partial_rotary_factor", 1.0)
-        head_dim = self.hparams["head_dim"]
-        rope_angles = int(partial * head_dim) // 2
-        n_pairs = head_dim // 2
-        assert 0 < rope_angles <= n_pairs
-        values = [1.0] * rope_angles + [1e30] * (n_pairs - rope_angles)
-        rope_freqs = torch.tensor(values, dtype=torch.float32)
-        yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), rope_freqs)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        # conv op requires 2d tensor
-        if "conv.conv" in name:
-            data_torch = data_torch.squeeze(1)
-
-        yield from super().modify_tensors(data_torch, name, bid)
-
-
-# d1 on LFM3-VL / LFM3-Omni: the stock weights, the d1 prompt read at its last token
-_D1_TYPES = {"lfm3_vl": gguf.DecisionType.LFM3_D1, "lfm3_omni": gguf.DecisionType.LFM3_D1_OMNI}
-
-
-def _is_lfm3_d1_checkpoint(dir_model: Path) -> bool:
-    if not (dir_model / "config.json").is_file():
-        return False
-    with open(dir_model / "config.json", encoding="utf-8") as f:
-        config = json.load(f)
-    return config.get("architectures") == ["D1Model"] and config.get("model_type") in _D1_TYPES
-
-
-@ModelBase.register_hparams_loader(_is_lfm3_d1_checkpoint)
-def _load_lfm3_d1_hparams(dir_model: Path) -> dict[str, Any]:
-    logger.info("gguf: detected LFM3 d1 checkpoint")
-    hparams = ModelBase.load_hparams(dir_model, False, guess=False)
-    hparams["d1_family"] = hparams["model_type"]
-    # the mmproj is the base model's, the text model writes the decision
-    hparams["architectures"] = ["Lfm3VlForConditionalGeneration" if hparams["model_type"] == "lfm3_vl" else "Lfm3OmniForConditionalGeneration"]
-    hparams["text_config"]["architectures"] = ["Lfm3D1Model"]
-    return hparams
-
-
-@ModelBase.register("Lfm3D1Model")
-@ModelBase.example("LiquidAI/d1-2B", "LiquidAI/d1-omni-2B")
-class LFM3D1Model(LFM3Model):
-    model_arch = gguf.MODEL_ARCH.LFM3
-
-    # how the checkpoint serves, hard-coded in the server: assert the checkpoint records the same
-    _served = {"d1_codes": 2, "d1_attention": "set"}
-
-    def set_vocab(self):
-        # the vocab without the speech task templates of LFM3-Audio/Omni, d1 only decides
-        TextModel.set_vocab(self)
-        # the d1 prompt of LFM2, the LFM3 template writes an empty thought before the answer
-        self.gguf_writer.add_chat_template([{"name": "systemone", "template": D1Model._systemone_template() + "<think></think>"}])
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-
-        for key, value in self._served.items():
-            if self.hparams.get(key) != value:
-                raise ValueError(f"unsupported d1 checkpoint: {key} = {self.hparams.get(key)!r}, expected {value!r}")
-
-        # a lone question reads the whole prompt both ways on the full attention layers
-        self.gguf_writer.add_causal_attention(False)
-        self.gguf_writer.add_decision_type(_D1_TYPES[self.hparams["d1_family"]])
-
-        # the option answers share one temperature
-        with open(self.dir_model / "calibration.json", encoding="utf-8") as f:
-            calibration = json.load(f)
-        if calibration.get("temperatures"):
-            raise ValueError("unsupported d1 calibration: per-bucket temperatures")
-        for name in ("noul", "choice", "score"):
-            self.gguf_writer.add_decision_temperature(name, calibration["default"])
-
-
-class LFM3VLVisionMixin(MmprojModel):
-    """LFM3-VL vision tower and projector, shared with LFM3-Omni."""
-
-    def _set_lfm3vl_gguf_parameters(self):
-        assert self.hparams_vision is not None
-        # clip.cpp hardcodes the GELU projector and its input norm eps
-        if (self.global_config["projector_hidden_act"] != "gelu" or not self.global_config["projector_bias"]
-                or self.global_config["projector_layer_norm_eps"] != 1e-6):
-            raise ValueError("unsupported projector_hidden_act / projector_bias / projector_layer_norm_eps")
-        # clip.cpp hardcodes the vision rope base, the RGB patchifier and the 3x3 spatial conv
-        if (self.hparams_vision["rope_parameters"]["rope_theta"] != 1e6 or self.hparams_vision["in_channels"] != 3
-                or self.hparams_vision["conv_kernel_size"] != 3):
-            raise ValueError("unsupported vision rope_theta / in_channels / conv_kernel_size")
-        self.gguf_writer.add_clip_vision_projector_type(gguf.VisionProjectorType.LFM3VL)
-        self.gguf_writer.add_vision_head_count_kv(self.hparams_vision["num_key_value_heads"])
-        self.gguf_writer.add_vision_spatial_merge_size(self.global_config["downsample_factor"])
-        self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams_vision["norm_eps"])
-        self.gguf_writer.add_vision_use_silu(True)
-        size = self.preprocessor_config["size"]
-        self.gguf_writer.add_vision_min_pixels(size["shortest_edge"])
-        self.gguf_writer.add_vision_max_pixels(size["longest_edge"])
-        # PIL resample enum, the torchvision backend has no lanczos
-        resample = self.preprocessor_config["resample"]
-        resize_algo = {2: "bilinear", 3: "bicubic"}.get(resample)
-        if resize_algo is None:
-            raise ValueError(f"unsupported resample: {resample}")
-        self.gguf_writer.add_vision_image_resize_algo(resize_algo)
-        self._check_video_config()
-
-    # mtmd_get_video_ts_style and the lfm3vl video frame sizing hardcode these, conversion fails on a mismatch
-    _video = {"fps": 2.0, "min_frames": 4, "temporal_patch_size": 2, "size": {"shortest_edge": 4096, "longest_edge": 25165824}}
-    _video_max_frames = 768
-
-    def _check_video_config(self):
-        import json
-        video = self.preprocessor_config.get("video_processor")
-        path = self.dir_model / "video_preprocessor_config.json"
-        if video is None and path.is_file():
-            with open(path, encoding="utf-8") as f:
-                video = json.load(f)
-        if video is None:
-            return
-        for k, v in {**self._video, "max_frames": self._video_max_frames}.items():
-            if video.get(k) != v:
-                raise ValueError(f"unsupported video {k}={video.get(k)!r}, expected {v!r}")
-
-    def tensor_force_quant(self, name, new_name, bid, n_dims):
-        if "patch_embd" in new_name:
-            return gguf.GGMLQuantizationType.F16
-        if "conv2d_spatial" in new_name:
-            # CUDA CONV_2D_DW requires an f32 kernel; the tensor is tiny
-            return gguf.GGMLQuantizationType.F32
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
-
-    def _lfm3vl_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name == "model.vision_tower.patchifier.weight":
-            # qwen2vl-style layout [channel, temporal, ph, pw], one conv kernel per temporal slice
-            assert self.hparams_vision is not None
-            t = self.hparams_vision["temporal_patch_size"]
-            p = self.hparams_vision["patch_size"]
-            n_embd = data_torch.shape[0]
-            w = data_torch.view(n_embd, 3, t, p, p)
-            yield ("v.patch_embd.weight", w[:, :, 0].contiguous())
-            if t == 2:
-                yield ("v.patch_embd.weight.1", w[:, :, 1].contiguous())
-            else:
-                assert t == 1, f"unsupported vision_temporal_patch_size {t}"
-            return
-        if name == "model.vision_tower.embedding_norm.weight":
-            yield ("v.embedding_norm.weight", data_torch)
-            return
-        if "conv.in_proj.weight" in name:
-            yield (f"v.blk.{bid}.conv2d_in_proj.weight", data_torch)
-            return
-        if "conv.spatial_conv.weight" in name:
-            yield (f"v.blk.{bid}.conv2d_spatial.weight", data_torch)
-            return
-        if "conv.out_proj.weight" in name:
-            yield (f"v.blk.{bid}.conv2d_out_proj.weight", data_torch)
-            return
-
-        # projector: input LayerNorm -> linear_1 -> GELU -> linear_2
-        if name.startswith("model.multi_modal_projector."):
-            layer, suffix = name.removeprefix("model.multi_modal_projector.").split(".")
-            yield ({"layer_norm": "mm.input_norm", "linear_1": "mm.0", "linear_2": "mm.2"}[layer] + "." + suffix, data_torch)
-            return
-
-        yield (self.map_tensor_name(name), data_torch)
-
-
-@ModelBase.register("Lfm3VlForConditionalGeneration")
-@ModelBase.example("LiquidAI/LFM3-VL-2B")
-class LFM3VLMmprojModel(LFM3VLVisionMixin):
-    has_vision_encoder = True
-    has_audio_encoder = False
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        assert self.hparams_vision is not None
-        # dynamic resolution, image_size is only set for MmprojModel compatibility
-        self.hparams_vision["image_size"] = 512
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        self._set_lfm3vl_gguf_parameters()
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.startswith(("model.vision_tower.", "model.multi_modal_projector.")):
-            yield from self._lfm3vl_tensors(data_torch, name, bid)
-
-
-def _is_lfm3_audio_checkpoint(dir_model: Path) -> bool:
-    if not (dir_model / "config.json").is_file():
-        return False
-    with open(dir_model / "config.json", encoding="utf-8") as f:
-        config = json.load(f)
-    return config.get("model_type") == "lfm3_audio" and "lfm" in config
-
-
-# read the release config.json as is: a transformers that knows lfm3_audio rewrites it into another layout
-@ModelBase.register_hparams_loader(_is_lfm3_audio_checkpoint)
-def _load_lfm3_audio_hparams(dir_model: Path) -> dict[str, Any]:
-    with open(dir_model / "config.json", encoding="utf-8") as f:
-        config = json.load(f)
-    config["text_config"] = config["lfm"]
-    return config
-
-
-@ModelBase.register("Lfm3AudioForConditionalGeneration")
-@ModelBase.example("LiquidAI/LFM3-Audio-2B")
-class LFM3AudioMmprojModel(MmprojModel):
-    has_vision_encoder = False
-    has_audio_encoder = True
-
-    # the downsampler has the LFM2-Audio pre-encoder layout: conv, dw conv, pw conv, dw conv, pw conv
-    _downsample_idx = {"conv_in": 0, "conv_mid": 2, "lin_mid": 3, "conv_end": 5, "lin_end": 6}
-
-    # audio output: MaskGIT depthformer (a.gen.code.*) and the codec decoder (a.gen.wav.*)
-    _depth_layer_map = {
-        "operator_norm":      gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_NORM,
-        "self_attn.q_proj":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_Q,
-        "self_attn.q_norm":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_Q_NORM,
-        "self_attn.k_proj":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_K,
-        "self_attn.k_norm":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_K_NORM,
-        "self_attn.v_proj":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_V,
-        "self_attn.o_proj":   gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_OUT,
-        "self_attn.gate_proj": gguf.MODEL_TENSOR.A_GEN_CODE_ATTN_GATE,
-        "ffn_norm":           gguf.MODEL_TENSOR.A_GEN_CODE_FFN_NORM,
-        "feed_forward.w1":    gguf.MODEL_TENSOR.A_GEN_CODE_FFN_GATE,
-        "feed_forward.w3":    gguf.MODEL_TENSOR.A_GEN_CODE_FFN_UP,
-        "feed_forward.w2":    gguf.MODEL_TENSOR.A_GEN_CODE_FFN_DOWN,
-    }
-    _codec_layer_map = {
-        "operator_norm":         gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_NORM,
-        "post_operator_norm":    gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_POST_NORM,
-        "self_attn.q_proj":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_Q,
-        "self_attn.q_norm":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_Q_NORM,
-        "self_attn.k_proj":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_K,
-        "self_attn.k_norm":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_K_NORM,
-        "self_attn.v_proj":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_V,
-        "self_attn.o_proj":      gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_OUT,
-        "self_attn.gate_proj":   gguf.MODEL_TENSOR.A_GEN_WAV_TFM_ATTN_GATE,
-        "conv.in_proj":          gguf.MODEL_TENSOR.A_GEN_WAV_TFM_CONV_PW1,
-        "conv.conv":             gguf.MODEL_TENSOR.A_GEN_WAV_TFM_CONV_DW,
-        "conv.out_proj":         gguf.MODEL_TENSOR.A_GEN_WAV_TFM_CONV_PW2,
-        "ffn_norm":              gguf.MODEL_TENSOR.A_GEN_WAV_TFM_FFN_NORM,
-        "post_ffn_norm":         gguf.MODEL_TENSOR.A_GEN_WAV_TFM_FFN_POST_NORM,
-        "feed_forward.w1":       gguf.MODEL_TENSOR.A_GEN_WAV_TFM_FFN_GATE,
-        "feed_forward.w3":       gguf.MODEL_TENSOR.A_GEN_WAV_TFM_FFN_UP,
-        "feed_forward.w2":       gguf.MODEL_TENSOR.A_GEN_WAV_TFM_FFN_DOWN,
-    }
-    # the codec graph in clip.cpp hardcodes the liquid-codec FSQ front end and decoder shape, conversion fails on a mismatch
-    _codec = {
-        "model_type": "liquid_codec", "fsq_levels": [8, 7, 6, 6], "n_q": 12, "quantizer_dim": 256,
-        "encoder_dim": 512, "stft_n_fft": 1280, "stft_hop_length": 320, "stft_win_length": 1280, "stft_frame_stack": 6,
-        "decoder_window": 30, "sampling_rate": 24000,
-    }
-    # the lfm3a pipeline in mtmd-helper-gen.cpp hardcodes these
-    _generation = {"codebooks": 12, "maskgit_schedule": [4, 4, 4], "interleaved_n_text": 6, "interleaved_n_audio": 12}
-
-    # speech-code sampling of the model card, generation_config.json overrides it
-    _audio_sampling = {
-        "audio_temperature": 1.0, "audio_top_k": 128, "audio_repetition_penalty": 1.1, "audio_repetition_window": 4096,
-    }
-
-    # mtmd_audio_preprocessor_lfm3a hardcodes this front end, conversion fails on a mismatch
-    _feature_extractor = {
-        "sampling_rate": 16000, "feature_size": 128, "n_fft": 400, "win_length": 400, "hop_length": 160,
-        "log_floor": -12.0, "normalization_offset": 4.0, "normalization_divisor": 4.0, "encoder_stride": 8,
-    }
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        # audio output is exported when the checkpoint ships the codec
-        self._audio_output = (self.dir_model / "audio_tokenizer").is_dir()
-        gen = self.get_generation_config()
-        self._n_codebooks: int = gen["codebooks"]
-        self._audio_vocab: int = gen["audio_vocab_size"]
-        self._depth_embd: dict[int, Tensor] = {}
-        self._depth_norm: dict[int, Tensor] = {}
-
-    def get_audio_config(self) -> dict[str, Any] | None:
-        enc = self.global_config["encoder"]
-        cfg = dict(enc["lfm"])
-        cfg["num_mel_bins"] = enc["dim_in"]
-        cfg["sliding_window"] = enc["sliding_window"]
-        return cfg
-
-    def get_generation_config(self) -> dict[str, Any]:
-        c = self.global_config
-        steps = c["maskgit_steps"]
-        return {
-            "codebooks": c["codebooks"], "audio_vocab_size": c["audio_vocab_size"],
-            "maskgit_schedule": [c["codebooks"] // steps] * steps,
-            "interleaved_n_text": c["interleaved_n_text"], "interleaved_n_audio": c["interleaved_n_audio"],
-        }
-
-    def get_depth_config(self) -> dict[str, Any]:
-        return self.global_config["depthformer"]
-
-    def set_gguf_parameters(self):
-        assert self.hparams_audio is not None
-        fe = self.preprocessor_config["feature_extractor"]
-        if fe["feature_extractor_type"] not in ("Lfm3AudioFeatureExtractor", "Lfm3OmniFeatureExtractor"):
-            raise ValueError(f"unsupported feature_extractor_type {fe['feature_extractor_type']!r}")
-        for k, v in self._feature_extractor.items():
-            if fe[k] != v:
-                raise ValueError(f"unsupported feature_extractor.{k}={fe[k]!r}, expected {v!r}")
-
-        # clip.cpp hardcodes the encoder rope base
-        if self.hparams_audio["rope_parameters"]["rope_theta"] != 1e6:
-            raise ValueError("unsupported audio encoder rope_theta")
-
-        self.hparams_audio["intermediate_size"] = lfm2_feed_forward_length(self.hparams_audio)
-        super().set_gguf_parameters()
-        self.gguf_writer.add_clip_audio_projector_type(gguf.VisionProjectorType.LFM3A)
-        self.gguf_writer.add_audio_head_count_kv(self.hparams_audio["num_key_value_heads"])
-        self.gguf_writer.add_audio_attention_layernorm_eps(self.hparams_audio["norm_eps"])
-        self.gguf_writer.add_audio_num_mel_bins(self.hparams_audio["num_mel_bins"])
-        self.gguf_writer.add_audio_window_size(self.hparams_audio["sliding_window"])
-
-        if not self._audio_output:
-            return
-
-        # audio generation: depthformer hparams, the codec ones are constant and live in clip.cpp
-        gen = self.get_generation_config()
-        for k, v in self._generation.items():
-            if gen[k] != v:
-                raise ValueError(f"unsupported {k}={gen[k]!r}, expected {v!r}")
-        depth = self.get_depth_config()
-        if depth.get("is_causal", False) or depth["use_sandwich_norm"] or set(depth["layer_types"]) != {"full_attention"}:
-            raise ValueError("unsupported depthformer, expected a bidirectional all-attention stack")
-        if depth["rope_parameters"]["full_attention"]["rope_theta"] != 1e6:
-            raise ValueError("unsupported depthformer rope_theta")
-        self.gguf_writer.add_clip_has_gen_audio_encoder(True)
-        self.gguf_writer.add_clip_gen_audio_projector_type(gguf.VisionProjectorType.LFM3A_GEN)
-        self.gguf_writer.add_gen_audio_projection_dim(self.n_embd_text)
-        self.gguf_writer.add_gen_audio_embedding_length(depth["hidden_size"])
-        self.gguf_writer.add_gen_audio_feed_forward_length(depth["intermediate_size"])
-        self.gguf_writer.add_gen_audio_block_count(depth["num_hidden_layers"])
-        self.gguf_writer.add_gen_audio_head_count(depth["num_attention_heads"])
-        self.gguf_writer.add_gen_audio_head_count_kv(depth["num_key_value_heads"])
-        self.gguf_writer.add_gen_audio_attention_layernorm_eps(depth["norm_eps"])
-
-        # sampling defaults of the speech codes, next to the text ones that become general.sampling.*
-        gen = gguf.Metadata.load_generation_config(self.dir_model)
-        for key, add in (("audio_temperature",        self.gguf_writer.add_gen_audio_sampling_temp),
-                         ("audio_top_k",              self.gguf_writer.add_gen_audio_sampling_top_k),
-                         ("audio_repetition_penalty", self.gguf_writer.add_gen_audio_sampling_penalty_repeat),
-                         ("audio_repetition_window",  self.gguf_writer.add_gen_audio_sampling_penalty_last_n)):
-            value = gen.get(key, self._audio_sampling.get(key))
-            if value is not None:
-                add(value)
-
-    def tensor_force_quant(self, name, new_name, bid, n_dims):
-        if new_name.startswith("a.conv1d.") or "conv_dw" in new_name or new_name.startswith("a.gen.wav.quant."):
-            return gguf.GGMLQuantizationType.F32
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
-
-    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
-        if self._audio_output:
-            yield from self._codec_tensors()
-
-    def _codec_tensors(self) -> Iterable[tuple[str, Tensor]]:
-        import json
-        from safetensors.torch import load_file
-
-        T = gguf.MODEL_TENSOR
-        codec_dir = self.dir_model / "audio_tokenizer"
-        with open(codec_dir / "config.json", encoding="utf-8") as f:
-            cfg = json.load(f)
-        for k, v in self._codec.items():
-            if cfg.get(k) != v:
-                raise ValueError(f"unsupported audio_tokenizer {k}={cfg.get(k)!r}, expected {v!r}")
-        if math.prod(cfg["fsq_levels"]) != self._audio_vocab - 1:
-            raise ValueError(f"audio_vocab_size {self._audio_vocab} does not match the codec, expected codebook size + 1")
-
-        state_dict = load_file(codec_dir / "model.safetensors")
-
-        # FSQ code table: index -> the len(levels) quantized values, see FiniteScalarQuantization._indices_to_codes
-        levels = torch.tensor(cfg["fsq_levels"], dtype=torch.int64)
-        basis = torch.cumprod(torch.cat([torch.ones(1, dtype=torch.int64), levels[:-1]]), dim=0)
-        half = levels // 2
-        idx = torch.arange(int(levels.prod()), dtype=torch.int64)[:, None]
-        codes = ((idx // basis) % levels - half) / half
-        yield (self.format_tensor_name(T.A_GEN_WAV_QUANT_FIRST_CB), codes.to(torch.float32))
-
-        # one projection over the concatenated code vectors of all codebooks, under quantizer.fsq. in newer exports
-        quantizer = "quantizer.fsq." if "quantizer.fsq.project_out.weight" in state_dict else "quantizer."
-        for suffix in ("weight", "bias"):
-            yield (self.format_tensor_name(T.A_GEN_WAV_QUANT_FIRST_OUT, suffix=f".{suffix}"), state_dict[f"{quantizer}project_out.{suffix}"])
-            yield (self.format_tensor_name(T.A_GEN_WAV_TFM_IN_PROJ, suffix=f".{suffix}"), state_dict[f"from_quantizer.{suffix}"])
-            yield (self.format_tensor_name(T.A_GEN_WAV_TFM_OUT_PROJ, suffix=f".{suffix}"), state_dict[f"decoder.lin.{suffix}"])
-        yield (self.format_tensor_name(T.A_GEN_WAV_TFM_OUTPUT_NORM), state_dict["decoder.lfm.embedding_norm.weight"])
-
-        for name, data_torch in state_dict.items():
-            if not name.startswith("decoder.lfm.layers."):
-                continue
-            bid_str, key = name.removeprefix("decoder.lfm.layers.").rsplit(".weight", 1)[0].split(".", 1)
-            if key == "conv.conv":
-                data_torch = data_torch.squeeze(1)
-            yield (self.format_tensor_name(self._codec_layer_map[key], int(bid_str)), data_torch)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.startswith("a.gen.wav."):
-            yield (name, data_torch)
-            return
-
-        if name.startswith(("depthformer.", "depth_", "codebook_embedding.", "mask_embedding")):
-            if self._audio_output:
-                yield from self._depthformer_tensor(data_torch, name, bid)
-            return
-
-        if not name.startswith(("encoder.", "mlp.")):
-            return
-
-        if name.startswith("encoder.downsample."):
-            layer, suffix = name.removeprefix("encoder.downsample.").split(".")
-            if layer == "lin_out":
-                if suffix == "weight":
-                    data_torch = data_torch.squeeze(-1)
-                yield (self.format_tensor_name(gguf.MODEL_TENSOR.A_ENC_OUT, suffix=f".{suffix}"), data_torch)
-                return
-            if suffix == "bias":
-                data_torch = data_torch[:, None, None]
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.A_ENC_CONV1D, self._downsample_idx[layer], f".{suffix}"), data_torch)
-            return
-
-        # conv op requires 2d tensor
-        if name.endswith("conv.conv.weight"):
-            data_torch = data_torch.squeeze(1)
-
-        yield (self.map_tensor_name(name), data_torch)
-
-    def _depthformer_tensor(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        T = gguf.MODEL_TENSOR
-
-        if name.startswith("depthformer.model.layers."):
-            key = name.removeprefix("depthformer.model.layers.").split(".", 1)[1].rsplit(".weight", 1)[0]
-            yield (self.format_tensor_name(self._depth_layer_map[key], bid), data_torch)
-            return
-        if name == "depthformer.model.embedding_norm.weight":
-            yield (self.format_tensor_name(T.A_GEN_CODE_OUTPUT_NORM), data_torch)
-            return
-        if name.startswith("depth_linear."):
-            yield (self.format_tensor_name(T.A_GEN_CODE_PROJ_IN, suffix="." + name.rsplit(".", 1)[1]), data_torch)
-            return
-        if name == "mask_embedding":
-            yield (self.format_tensor_name(T.A_GEN_CODE_MASK_EMBD, suffix=""), data_torch)
-            return
-        if name == "codebook_embedding.weight":
-            assert data_torch.shape[0] == self._n_codebooks * self._audio_vocab
-            yield (self.format_tensor_name(T.A_GEN_CODE_OUT_EMBD), data_torch)
-            return
-
-        # per-codebook heads: the embedding table is tied to to_logits, merge the 12 tables into one tensor
-        assert name.startswith("depth_embeddings.")
-        idx = int(name.split(".")[1])
-        key = name.split(".", 2)[2]
-        if key == "embedding.weight":
-            self._depth_embd[idx] = data_torch
-            if len(self._depth_embd) == self._n_codebooks:
-                yield (self.format_tensor_name(T.A_GEN_CODE_EMBD), torch.stack([self._depth_embd[i] for i in range(self._n_codebooks)], dim=0))
-        elif key == "embedding_norm.weight":
-            self._depth_norm[idx] = data_torch
-            if len(self._depth_norm) == self._n_codebooks:
-                yield (self.format_tensor_name(T.A_GEN_CODE_HEAD_NORM), torch.stack([self._depth_norm[i] for i in range(self._n_codebooks)], dim=0))
-        elif key == "to_logits.weight":
-            if not torch.equal(data_torch, self._depth_embd[idx]):
-                raise ValueError(f"depth_embeddings.{idx} is not tied")
-        else:
-            raise ValueError(f"unexpected tensor {name}")
-
-
-@ModelBase.register("Lfm3OmniForConditionalGeneration")
-@ModelBase.example("LiquidAI/LFM3-Omni-2B")
-class LFM3OmniMmprojModel(LFM3VLVisionMixin, LFM3AudioMmprojModel):
-    has_vision_encoder = True
-    has_audio_encoder = True
-
-    _audio_sampling = {"audio_temperature": 0.4, "audio_repetition_penalty": 1.2, "audio_repetition_window": 64}
-    _video_max_frames = 384
-
-    # the audio side is the LFM3-Audio model under the HF-native names
-    _audio_renames = (
-        ("model.audio_encoder.lfm.", "encoder.lfm.lfm."),
-        ("model.audio_encoder.", "encoder."),
-        ("model.audio_projector.layer_norm.", "mlp.model.0."),
-        ("model.audio_projector.linear_1.", "mlp.model.1."),
-        ("model.audio_projector.linear_2.", "mlp.model.3."),
-        ("model.depth_decoder.model.", "depthformer.model."),
-        ("model.depth_decoder.", ""),
-        ("model.codebook_embedding.", "codebook_embedding."),
-    )
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        assert self.hparams_vision is not None
-        # dynamic resolution, image_size is only set for MmprojModel compatibility
-        self.hparams_vision["image_size"] = 512
-
-    def get_audio_config(self) -> dict[str, Any] | None:
-        enc = self.global_config["audio_encoder_config"]
-        cfg = dict(enc["text_config"])
-        cfg["num_mel_bins"] = enc["num_mel_bins"]
-        cfg["sliding_window"] = enc["sliding_window"]
-        return cfg
-
-    def get_generation_config(self) -> dict[str, Any]:
-        c = self.global_config
-        return {
-            "codebooks": c["num_codebooks"], "audio_vocab_size": c["codebook_size"], "maskgit_schedule": c["maskgit_schedule"],
-            "interleaved_n_text": c["interleaved_n_text"], "interleaved_n_audio": c["interleaved_n_audio"],
-        }
-
-    def get_depth_config(self) -> dict[str, Any]:
-        return self.global_config["depth_config"]
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        self._set_lfm3vl_gguf_parameters()
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.startswith(("model.vision_tower.", "model.multi_modal_projector.")):
-            yield from self._lfm3vl_tensors(data_torch, name, bid)
-            return
-        for old, new in self._audio_renames:
-            if name.startswith(old):
-                yield from super().modify_tensors(data_torch, new + name.removeprefix(old), bid)
-                return
-        yield from super().modify_tensors(data_torch, name, bid)
@@ -1,184 +0,0 @@
-from __future__ import annotations
-
-from typing import Iterable, TYPE_CHECKING
-
-import torch
-
-if TYPE_CHECKING:
-    from torch import Tensor
-
-from .base import ModelBase, TextModel, gguf
-
-
-@ModelBase.register("LimiteForCausalLM")
-@ModelBase.example("paradigma-inc/limite-1b-violetto")
-class LimiteModel(TextModel):
-    model_arch = gguf.MODEL_ARCH.LIMITE
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        hp = self.hparams
-
-        expected = {
-            "pos_mode": "rope",
-            "qk_norm": "rms_pre_rope",
-            "rope_style": "interleaved_pairs_odd_lane_sign_flip",
-            "rope_per_layer": False,
-            "rms_norm_has_weight": False,
-            "rms_norm_eps_mode": "torch_finfo_default",
-            "mlp_type": "swiglu",
-            "attn_gate_applied": "per_head_before_o_proj",
-            "ve_head_slice": "first_num_key_value_heads",
-            "ve_applied_before_qk_norm": True,
-            "global_nope": True,
-            "global_window": -1,
-            "mudd_r_site": "resid",
-            "tie_word_embeddings": True,
-            "final_softcap": 0.0,
-        }
-        for key, value in expected.items():
-            if hp.get(key) != value:
-                raise NotImplementedError(f"Limite: unsupported {key}={hp.get(key)!r}, expected {value!r}")
-        if hp["softcap_logits"]["kind"] != "sigmoid":
-            raise NotImplementedError(f"Limite: unsupported softcap_logits kind {hp['softcap_logits']['kind']!r}")
-
-        self.n_head    = hp["num_attention_heads"]
-        self.n_head_kv = hp["num_key_value_heads"]
-        self.head_dim  = hp["head_dim"]
-        self.n_pairs   = hp["rope_n_pairs"]
-
-        self.mudd_taps: dict[int, list[int]] = {}
-        if hp.get("mudd"):
-            if not hp.get("mudd_mlp"):
-                raise NotImplementedError("Limite: mudd without mudd_mlp is not supported")
-            self.mudd_taps = {int(il): [int(t) for t in taps] for il, taps in hp["mudd_tap_idx"].items()}
-            assert all(len(taps) == hp["mudd_taps"] for taps in self.mudd_taps.values())
-
-    def set_vocab(self):
-        self._set_vocab_gpt2()
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        hp = self.hparams
-
-        self.gguf_writer.add_key_length(self.head_dim)
-        self.gguf_writer.add_value_length(self.head_dim)
-
-        # torch rms_norm with eps=None uses the fp32 epsilon, even for bf16 inputs
-        self.gguf_writer.add_layer_norm_rms_eps(torch.finfo(torch.float32).eps)
-        self.gguf_writer.add_attention_scale(hp["attention_softmax_scale"])
-        self.gguf_writer.add_xsa_eps(hp["xsa_normalize_eps"] if hp.get("xsa") else 0.0)
-
-        # HF freq_i = base^(-i/(n_pairs-1)), ggml uses freq_base^(-i/n_pairs)
-        self.gguf_writer.add_rope_dimension_count(2 * self.n_pairs)
-        self.gguf_writer.add_rope_freq_base(hp["rope_base_local"] ** (self.n_pairs / (self.n_pairs - 1)))
-
-        # HF keeps keys k >= q - window, so a window of W spans W + 1 tokens
-        global_layers = set(hp["global_layers"])
-        self.gguf_writer.add_sliding_window(hp["sliding_window"] + 1)
-        self.gguf_writer.add_sliding_window_pattern([il not in global_layers for il in range(self.block_count)])
-
-        cap = hp["softcap_logits"]
-        self.gguf_writer.add_final_logit_sigmoid_capping(cap["a"], cap["b"], cap["c"])
-
-        if self.mudd_taps:
-            n_taps = hp["mudd_taps"]
-            tap_indices = [-1] * (self.block_count * n_taps)
-            for il, taps in self.mudd_taps.items():
-                tap_indices[il * n_taps:(il + 1) * n_taps] = taps
-            self.gguf_writer.add_mudd_feed_forward_length(hp["mudd_inter"])
-            self.gguf_writer.add_mudd_tap_count(n_taps)
-            self.gguf_writer.add_mudd_tap_indices(tap_indices)
-
-    def tensor_force_quant(self, name, new_name, bid, n_dims):
-        # tiny mixing weights, kept at full precision as in the reference
-        # the mudd tensors come from a global source tensor, so take the block id from the new name
-        if new_name.startswith("blk."):
-            bid = int(new_name.split(".")[1])
-        for key in (gguf.MODEL_TENSOR.ATTN_GATE, gguf.MODEL_TENSOR.ATTN_VE_GATE, gguf.MODEL_TENSOR.MUDD_DOWN, gguf.MODEL_TENSOR.MUDD_UP):
-            if self.match_model_tensor_name(new_name, key, bid):
-                return gguf.GGMLQuantizationType.F32
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
-
-    def _swap_rope_pairs(self, w: Tensor, n_head: int) -> Tensor:
-        # HF rotates each pair by -theta; swapping the pair lanes gives the ggml +theta rotation.
-        # q and k are both swapped, so the attention scores do not change.
-        idx = torch.arange(self.head_dim)
-        idx[:2 * self.n_pairs] = idx[:2 * self.n_pairs].view(-1, 2).flip(-1).reshape(-1)
-        w = w.reshape(n_head, self.head_dim, -1)
-        return w[:, idx, :].reshape(n_head * self.head_dim, -1)
-
-    def _pad_cols(self, w: Tensor) -> Tensor:
-        # gates read only the first channels of the input, pad them to n_embd
-        return torch.nn.functional.pad(w, (0, self.hparams["hidden_size"] - w.shape[-1]))
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        hp = self.hparams
-        T = gguf.MODEL_TENSOR
-
-        def load(n: str) -> Tensor:
-            return self.model_tensors[n]().float()
-
-        if name == "model.value_embeds.weight":
-            ve = data_torch.reshape(data_torch.shape[0], -1, hp["ve_dim"])[:, :self.n_head_kv]
-            ve = torch.nn.functional.pad(ve, (0, self.head_dim - hp["ve_dim"]))
-            yield self.format_tensor_name(T.VALUE_EMBD), ve.reshape(ve.shape[0], -1) * hp["ve_gate_scale"]
-            return
-
-        if name == "model.mudd.dense1":
-            for il in self.mudd_taps:
-                yield self.format_tensor_name(T.MUDD_DOWN, il), data_torch
-            return
-        if name == "model.mudd.dense2":
-            # rows [0, n) mix the attention input, rows [n, 2n) mix the residual
-            dense2_mlp = load("model.mudd.dense2_mlp")
-            bias       = load("model.mudd.bias")
-            bias_mlp   = load("model.mudd.bias_mlp")
-            for il, taps in self.mudd_taps.items():
-                n = len(taps)
-                w = torch.cat([data_torch[il, :n], dense2_mlp[il, :n]], 0)
-                b = torch.cat([bias[il, :n], bias_mlp[il, :n]], 0)
-                yield self.format_tensor_name(T.MUDD_UP, il), w
-                yield self.format_tensor_name(T.MUDD_UP, il, ".bias"), b
-            return
-        if name.startswith("model.mudd."):
-            return
-
-        if bid is None:
-            yield from super().modify_tensors(data_torch, name, bid)
-            return
-
-        prefix = f"model.layers.{bid}."
-        name = name.removeprefix(prefix)
-
-        if name in ("self_attn.q_proj.weight", "self_attn.k_proj.weight", "self_attn.v_proj.weight"):
-            data_torch = data_torch * load(prefix + "self_attn.qkv_scale")
-            if name == "self_attn.q_proj.weight":
-                data_torch = self._swap_rope_pairs(data_torch, self.n_head)
-            elif name == "self_attn.k_proj.weight":
-                data_torch = self._swap_rope_pairs(data_torch, self.n_head_kv)
-        elif name == "self_attn.o_proj.weight":
-            data_torch = data_torch * load(prefix + "self_attn.o_scale") * load(prefix + "post_lambda_attn") * hp["attn_gate_scale"]
-        elif name == "mlp.down_proj.weight":
-            data_torch = data_torch * load(prefix + "post_lambda_mlp")
-        elif name == "self_attn.attn_gate":
-            yield self.format_tensor_name(T.ATTN_GATE, bid), self._pad_cols(data_torch)
-            return
-        elif name == "self_attn.ve_gate":
-            yield self.format_tensor_name(T.ATTN_VE_GATE, bid), self._pad_cols(data_torch[:self.n_head_kv])
-            return
-        elif name == "self_attn.xsa_alpha":
-            if hp.get("xsa") and bid in hp["xsa_layers"]:
-                yield self.format_tensor_name(T.ATTN_XSA_ALPHA, bid), torch.tanh(data_torch)
-            return
-        elif name == "resid_lambda_attn":
-            yield self.format_tensor_name(T.ATTN_RESID_SCALE, bid), data_torch.reshape(1)
-            return
-        elif name == "resid_lambda_mlp":
-            yield self.format_tensor_name(T.FFN_RESID_SCALE, bid), data_torch.reshape(1)
-            return
-        elif name in ("self_attn.qkv_scale", "self_attn.o_scale", "post_lambda_attn", "post_lambda_mlp"):
-            # folded into the projections above
-            return
-
-        yield from super().modify_tensors(data_torch, prefix + name, bid)
@@ -272,18 +272,3 @@ class MiniCPMV4_7VisionModel(MiniCPMV4_6VisionModel):
            yield self.format_tensor_name(gguf.MODEL_TENSOR.V_TOK_EMBD_SEP, suffix=""), data_torch[ids]
            return
        yield from super().modify_tensors(data_torch, name, bid)
-
-
-# MiniCPM-OCR 1.0: the 4.6 stack with model_type minicpm_ocr.
-
-@ModelBase.register("MiniCPMOcrForConditionalGeneration")
-@ModelBase.example("openbmb/MiniCPM-OCR-1.0")
-class MiniCPMOCR1_0TextModel(MiniCPMV4_6TextModel):
-    model_arch = gguf.MODEL_ARCH.QWEN35
-    no_mtp = True
-
-
-@ModelBase.register("MiniCPMOcrForConditionalGeneration")
-@ModelBase.example("openbmb/MiniCPM-OCR-1.0")
-class MiniCPMOCR1_0VisionModel(MiniCPMV4_6VisionModel):
-    projector_type = gguf.VisionProjectorType.MINICPM_OCR_1_0
@@ -39,11 +39,6 @@ class Mistral3Model(TextModel):
            self.gguf_writer.add_add_bos_token(False)
            self.gguf_writer.add_add_eos_token(False)

-        def get_vocab_base_pre(self, tokenizer) -> str:
-            if self.origin_hf_arch == "Mistral3Model":
-                return "nemotron-3.5-embed"
-            return super().get_vocab_base_pre(tokenizer)
-
        def set_gguf_parameters(self):
            super().set_gguf_parameters()
            self._try_set_pooling_type()
@@ -1,217 +0,0 @@
-from __future__ import annotations
-
-from typing import Callable, Iterable, TYPE_CHECKING
-
-import torch
-
-if TYPE_CHECKING:
-    from torch import Tensor
-
-from .base import ModelBase, gguf
-
-from .mistral import MistralModel, MistralMoeModel
-
-
-@ModelBase.register("MistralLarge4ForConditionalGeneration")
-class MistralLarge4Model(MistralMoeModel):
-    model_arch = gguf.MODEL_ARCH.MISTRAL_LARGE4
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        n_layer = self.hparams["num_hidden_layers"]
-        dsa = self.hparams["dsa"]
-
-        def expand(pattern: str) -> list[str]:
-            values = pattern.split(",")
-            return [values[i % len(values)] for i in range(n_layer)]
-
-        self.layer_has_rope = [v == "rope" for v in expand(self.hparams.get("nope") or "rope")]
-        self.layer_indexer_full = [v == "full" for v in expand(dsa.get("sharing") or "full")]
-
-        kv_sharing = self.hparams.get("kv_sharing") or {}
-        if "source_layers_str" in kv_sharing:
-            self.layer_kv_source = [int(v) for v in kv_sharing["source_layers_str"].split(",")]
-        else:
-            assert not kv_sharing, f"unsupported kv_sharing config: {kv_sharing}"
-            self.layer_kv_source = list(range(n_layer))
-        assert len(self.layer_kv_source) == n_layer
-
-        # note: the rope dim of the indexer must match the main attention
-        assert dsa.get("rope_head_dim", -1) in (-1, self.hparams["qk_rope_head_dim"])
-        assert dsa.get("match_nope", False), "indexer must follow the NoPE pattern of the main attention"
-        assert dsa.get("ssmax_scaling", False), "indexer without SSMax is not supported"
-        assert self.hparams.get("q_lora_rank") is None
-        assert not self.hparams["yarn"].get("apply_scale", True), "only YaRN without magnitude scaling is supported"
-
-    def dequant_model(self):
-        quant_config = self.hparams.get("quantization")
-        if quant_config is not None:
-            assert quant_config["qformat_weight"] == "fp8_e4m3"
-            assert quant_config.get("qscheme_weight") == "BLOCK"
-            self.hparams["quantization_config"] = {
-                "quant_method": "fp8",
-                "weight_block_size": [quant_config["block_size_out"], quant_config["block_size_in"]],
-            }
-        return super().dequant_model()
-
-    def set_gguf_parameters(self):
-        # skip the MistralMoeModel overrides, they apply the deepseek2 YaRN scaling
-        super(MistralMoeModel, self).set_gguf_parameters()
-        MistralModel.set_mistral_config(self.gguf_writer, self.hparams)
-
-        dsa = self.hparams["dsa"]
-        self.gguf_writer.add_indexer_head_count(dsa["n_heads"])
-        self.gguf_writer.add_indexer_key_length(dsa["head_dim"])
-        self.gguf_writer.add_indexer_top_k(dsa["topk"])
-        self.gguf_writer.add_indexer_types(self.layer_indexer_full)
-        self.gguf_writer.add_layer_norm_eps(dsa.get("norm_eps", 1e-6))
-
-        self.gguf_writer.add_rope_pattern(self.layer_has_rope)
-        self.gguf_writer.add_kv_source_map(self.layer_kv_source)
-        self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SOFTMAX)
-        self.gguf_writer.add_final_logit_softcapping(self.hparams["softmax_tanh_gating"])
-
-    @classmethod
-    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
-        name, gen = item
-
-        # exported separately with --dspark
-        if name.startswith("mtp_"):
-            return None
-
-        # experts need HF names for the merging logic in DeepseekV2Model
-        # note: bare "layers.N.experts" comes from the NVFP4 quantized_layers config
-        if ".experts." in name or name.endswith(".experts"):
-            name = "model." + name.replace(".experts", ".mlp.experts")
-            name = name.replace(".w1.", ".gate_proj.").replace(".w2.", ".down_proj.").replace(".w3.", ".up_proj.")
-
-        # skip the MistralMoeModel renaming, the qscale names are needed for block FP8 dequant
-        return super(MistralMoeModel, cls).filter_tensors((name, gen))
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name == "lerope.alpha":
-            # LeRoPE scales inv_freq by exp(alpha), ggml divides theta by the freq factors
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), torch.exp(-data_torch.float()))
-            return
-
-        if bid is not None and name.endswith((".ssmax_scaling.s", ".ssmax_scaling.b")):
-            tensor = gguf.MODEL_TENSOR.INDEXER_SSMAX if ".indexer." in name else gguf.MODEL_TENSOR.ATTN_SSMAX
-            suffix = ".weight" if name.endswith(".s") else ".bias"
-            yield (self.format_tensor_name(tensor, bid, suffix), data_torch.float())
-            return
-
-        if bid is not None and ".indexer.q_input_norm." in name:
-            suffix = ".weight" if name.endswith(".weight") else ".bias"
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.INDEXER_Q_NORM, bid, suffix), data_torch)
-            return
-
-        if bid is not None and ".indexer." in name:
-            name = "model." + name.replace(".attention.", ".self_attn.")
-
-        # NoPE layers have a larger non-rotary head, so wkv_b is split per layer
-        if bid is not None and name.endswith(".wkv_b.weight"):
-            n_head = self.hparams["num_attention_heads"]
-            v_head_dim = self.hparams["v_head_dim"]
-            qk_nope_head_dim = self.hparams["qk_nope_head_dim"]
-            if not self.layer_has_rope[bid]:
-                qk_nope_head_dim += self.hparams["qk_rope_head_dim"]
-
-            assert data_torch.shape[0] == n_head * (qk_nope_head_dim + v_head_dim)
-
-            kv_b = data_torch.view(n_head, qk_nope_head_dim + v_head_dim, data_torch.shape[-1])
-            k_b, v_b = torch.split(kv_b, [qk_nope_head_dim, v_head_dim], dim=1)
-            k_b = k_b.transpose(1, 2)
-
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_K_B, bid), k_b)
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_V_B, bid), v_b)
-            return
-
-        yield from super().modify_tensors(data_torch, name, bid)
-
-
-class MistralLarge4DSparkModel(MistralLarge4Model):
-    """Mistral Large 4 DSpark draft (mtp_dspark.*) as a DFLASH GGUF, export with --dspark"""
-    model_arch = gguf.MODEL_ARCH.DFLASH
-
-    _DSPARK_ROOT: dict[str, gguf.MODEL_TENSOR] = {
-        "backbone.target_projection.weight": gguf.MODEL_TENSOR.FC,
-        "backbone.target_norm.weight":       gguf.MODEL_TENSOR.ENC_OUTPUT_NORM,
-        "backbone.final_norm.weight":        gguf.MODEL_TENSOR.OUTPUT_NORM,
-        "backbone.mask_embedding":           gguf.MODEL_TENSOR.DSPARK_MASK_EMBD,
-        "autoregressive_head.embedding.weight": gguf.MODEL_TENSOR.DSPARK_MARKOV_W1,
-        "autoregressive_head.output.weight":    gguf.MODEL_TENSOR.DSPARK_MARKOV_W2,
-        "autoregressive_head.joint_proj.weight": gguf.MODEL_TENSOR.DSPARK_JOINT_PROJ,
-        "autoregressive_head.joint_proj.bias":   gguf.MODEL_TENSOR.DSPARK_JOINT_PROJ,
-    }
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        mtp = self.hparams.get("mtp") or {}
-        dspark = mtp.get("dspark")
-        if dspark is None:
-            raise ValueError("Mistral Large 4 DSpark export needs the mtp.dspark config")
-        assert dspark.get("head_type") == "rnn", f"unsupported DSpark head type: {dspark.get('head_type')}"
-        assert dspark.get("draft_input") == "position_specific", f"unsupported DSpark draft input: {dspark.get('draft_input')}"
-
-        windows = {int(v) for v in (mtp.get("ragged_attention") or "").split(",") if v.strip()}
-        assert len(windows) == 1, f"DSpark draft layers must share one sliding window, got {mtp.get('ragged_attention')}"
-
-        self._n_stage = int(mtp["n_layers"])
-        self._depth = int(mtp["depth"])
-        self._window = windows.pop()
-        self._target_layers = [int(v) for v in mtp["target_activation_layers"].split(",")]
-
-        self.layer_has_rope = [True] * self._n_stage
-
-    @classmethod
-    def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
-        name, gen = item
-        if not name.startswith("mtp_dspark."):
-            return None
-        name = name.removeprefix("mtp_dspark.")
-        if name.startswith("backbone.layers."):
-            return name.removeprefix("backbone."), gen
-        return f"dspark.{name}", gen
-
-    def set_vocab(self):
-        super().set_vocab()
-        # block uses mask_embd, mask token id is not used
-        self.gguf_writer.add_mask_token_id(0)
-
-    def set_gguf_parameters(self):
-        hp = self.hparams
-        w = self.gguf_writer
-        w.add_block_count(self._n_stage)
-        w.add_context_length(hp["max_position_embeddings"])
-        w.add_embedding_length(hp["hidden_size"])
-        w.add_feed_forward_length(hp["intermediate_size"])
-        w.add_head_count(hp["num_attention_heads"])
-        w.add_head_count_kv(1)
-        w.add_key_length(hp["kv_lora_rank"] + hp["qk_rope_head_dim"])
-        w.add_value_length(hp["kv_lora_rank"])
-        w.add_key_length_mla(hp["qk_nope_head_dim"] + hp["qk_rope_head_dim"])
-        w.add_value_length_mla(hp["v_head_dim"])
-        w.add_kv_lora_rank(hp["kv_lora_rank"])
-        w.add_rope_dimension_count(hp["qk_rope_head_dim"])
-        w.add_rope_freq_base(float(hp["rope_theta"]))
-        w.add_layer_norm_rms_eps(hp["rms_norm_eps"])
-        MistralModel.set_mistral_config(w, hp)
-        w.add_sliding_window(self._window)
-        w.add_final_logit_softcapping(hp["softmax_tanh_gating"])
-        w.add_block_size(self._depth)
-        # draft reads layer output = input of next layer
-        w.add_target_layers([layer + 1 for layer in self._target_layers])
-        w.add_sample_from_anchor(True)
-        w.add_has_confidence_head(False)
-        w.add_file_type(self.ftype)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name.startswith("dspark."):
-            key = name.removeprefix("dspark.")
-            if key not in self._DSPARK_ROOT:
-                raise ValueError(f"Unexpected DSpark tensor {name!r}")
-            suffix = ".bias" if key.endswith(".bias") else ".weight"
-            yield (self.format_tensor_name(self._DSPARK_ROOT[key], None, suffix), data_torch)
-            return
-
-        yield from super().modify_tensors(data_torch, name, bid)
@@ -1,6 +1,5 @@
 from __future__ import annotations

-import re
 from typing import Any, Callable, Iterable, Sequence, TYPE_CHECKING

 import numpy as np
@@ -368,21 +367,9 @@ class NemotronModel(TextModel):
    model_arch = gguf.MODEL_ARCH.NEMOTRON

    def set_vocab(self):
-        if (self.dir_model / "tokenizer.model").exists():
-            self._set_vocab_sentencepiece()
-            self.gguf_writer.add_pad_token_id(0)
-            self.gguf_writer.add_unk_token_id(1)
-        else:
-            # SPM-style BPE (Metaspace + byte fallback): mark <0xXX> tokens as BYTE so they decode to raw bytes
-            tokens, toktypes, tokpre = self.get_vocab_base()
-            toktypes = [gguf.TokenType.BYTE if re.fullmatch(r"<0x[0-9A-F]{2}>", tok) else typ for tok, typ in zip(tokens, toktypes)]
-            self.gguf_writer.add_tokenizer_model("gpt2")
-            self.gguf_writer.add_tokenizer_pre(tokpre)
-            self.gguf_writer.add_token_list(tokens)
-            self.gguf_writer.add_token_types(toktypes)
-
-            special_vocab = gguf.SpecialVocab(self.dir_model, load_merges=True)
-            special_vocab.add_to_gguf(self.gguf_writer)
+        self._set_vocab_sentencepiece()
+        self.gguf_writer.add_pad_token_id(0)
+        self.gguf_writer.add_unk_token_id(1)

    def set_gguf_parameters(self):
        super().set_gguf_parameters()
@@ -396,10 +383,7 @@ class NemotronModel(TextModel):
        rot_pct = self.rope_parameters["partial_rotary_factor"]
        n_embd = self.find_hparam(["hidden_size", "n_embd"])
        n_head = self.find_hparam(["num_attention_heads", "n_head"])
-        # Use explicit head_dim when available; fallback keeps standard Nemotron working.
-        # Without this, models where n_head * head_dim != n_embd (e.g. 48 heads, head_dim=128) produce the wrong rope dimension count.
-        head_dim = self.hparams.get("head_dim") or n_embd // n_head
-        self.gguf_writer.add_rope_dimension_count(int(rot_pct * head_dim))
+        self.gguf_writer.add_rope_dimension_count(int(rot_pct * n_embd) // n_head)

        # * RopeScaling for Nemotron
        factor = self.hparams.get("factor") or self.rope_parameters.get("factor")
@@ -433,11 +417,6 @@ class NemotronHModel(GraniteHybridModel):
    _ATTN_LAYER_TYPES = {"attention", "full_attention"}
    _MLP_LAYER_TYPES = {"moe"}

-    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
-        if ".mixer.experts." in name:
-            name = name.replace("backbone.", "model.").replace(".mixer.experts.", ".mlp.experts.")
-        return super()._map_fp8_weight_names(name)
-
    def __init__(self, *args, **kwargs):
        # We have to determine the correct model architecture (MoE vs non-MoE) before
        # calling the parent __init__. This is because the parent constructor
@@ -83,15 +83,6 @@ class Qwen2Model(TextModel):
 class Qwen2MoeModel(TextModel):
    model_arch = gguf.MODEL_ARCH.QWEN2MOE

-    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
-        if name.removesuffix(".weight").endswith(".mlp.experts.gate_up_proj"):
-            bid = next(int(part) for part in self.map_tensor_name(name).split(".") if part.isdecimal())
-            return (
-                self.format_tensor_name(gguf.MODEL_TENSOR.FFN_GATE_EXP, bid),
-                self.format_tensor_name(gguf.MODEL_TENSOR.FFN_UP_EXP, bid),
-            )
-        return super()._map_fp8_weight_names(name)
-
    def set_gguf_parameters(self):
        super().set_gguf_parameters()
        if (moe_intermediate_size := self.hparams.get("moe_intermediate_size")) is not None:
@@ -391,16 +382,6 @@ class _QwenMtpMixin:
 class Qwen3NextModel(_QwenMtpMixin, Qwen2MoeModel):
    model_arch = gguf.MODEL_ARCH.QWEN3NEXT

-    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
-        if name.endswith(".linear_attn.in_proj_qkvz.weight"):
-            bid = next(int(part) for part in self.map_tensor_name(name).split(".") if part.isdecimal())
-            # Both split projections keep the original per-tensor scales.
-            return (
-                self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_QKV, bid),
-                self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_GATE, bid),
-            )
-        return super()._map_fp8_weight_names(name)
-
    def set_gguf_parameters(self):
        super().set_gguf_parameters()
        self.gguf_writer.add_ssm_conv_kernel(self.hparams["linear_conv_kernel_dim"])
@@ -709,52 +690,6 @@ class Qwen3_5TextModel(_Qwen35MRopeMixin, _LinearAttentionVReorderBase):
        self._try_set_pooling_type()


-@ModelBase.register("MiniCPMOCRForSequenceClassification")
-class MiniCPMOCRSeqClsModel(Qwen3_5TextModel):
-    # OCR consistency judge: Qwen3.5 VL trunk with a value head read at the last token
-    # the class scores are [inconsistent, consistent], the softmax temperature is 1
-    model_arch = gguf.MODEL_ARCH.QWEN35
-    no_mtp = True  # the checkpoint has no MTP head
-
-    CLASS_LABELS = ["inconsistent", "consistent"]
-
-    def set_vocab(self):
-        super().set_vocab()
-        self.gguf_writer.add_chat_template([{"name": "systemone", "template": self._systemone_template()}])
-
-    def _systemone_template(self) -> str:
-        # the prompt of the training data, the two images are the original page and the one
-        # rendered from the OCR text; the task is fixed, there is nothing to fill in
-        # the trailing newlines are in an expression, so the template does not drop them
-        return (
-            "<|im_start|>user\n{{ images[0] }}\nOCR文本：\n{{ state }}\n{{ images[1] }}\n"
-            "第一张图是原图，第二张图是上述 OCR 文本渲染得到的图片。\n"
-            "请判断该 OCR 文本及其对应渲染图是否与原图内容一致。\n"
-            "这是一个二分类任务：一致标为 1，不一致标为 0。"
-            "{{ '<|im_end|>\\n<|im_start|>assistant\\n<think>\\n\\n</think>\\n\\n' }}"
-        )
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-
-        # the config.json of the checkpoint keeps the one of the base model, it has no num_labels / id2label
-        if "score.weight" not in self.model_tensors:
-            raise ValueError("the checkpoint has no score.weight classification head")
-
-        self.gguf_writer.add_pooling_type(gguf.PoolingType.LAST)
-        self.gguf_writer.add_embedding_length_out(len(self.CLASS_LABELS))
-        self.gguf_writer.add_classifier_output_labels(self.CLASS_LABELS)
-        self.gguf_writer.add_decision_type(gguf.DecisionType.MINICPM_OCR_JEV)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        if name == "score.weight":
-            if data_torch.shape[0] != len(self.CLASS_LABELS):
-                raise ValueError(f"score.weight must have {len(self.CLASS_LABELS)} rows, got {data_torch.shape[0]}")
-            yield (gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.CLS_OUT] + ".weight", data_torch)
-            return
-        yield from super().modify_tensors(data_torch, name, bid)
-
-
 def _is_openjev_checkpoint(dir_model: Path) -> bool:
    return (dir_model / "helper" / "shim.py").is_file() and (dir_model / "config.json").is_file()

@@ -144,16 +144,6 @@ class Qwen3VLVisionModel(MmprojModel):
        yield from MmprojModel.modify_tensors(self, data_torch, name, bid)


-@ModelBase.register("MiniCPMOCRForSequenceClassification")
-class MiniCPMOCRSeqClsVisionModel(Qwen3VLVisionModel):
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        # trained with the image budget of its processor, wider than the mtmd default of Qwen-VL
-        size = self.preprocessor_config["size"]
-        self.gguf_writer.add_vision_min_pixels(int(size["shortest_edge"]))
-        self.gguf_writer.add_vision_max_pixels(int(size["longest_edge"]))
-
-
@ModelBase.register("Qwen3OmniMoeForConditionalGeneration")
@ModelBase.example("Qwen/Qwen3-Omni-30B-A3B-Instruct")
 class Qwen3OmniMmprojModel(Qwen3VLVisionModel, Qwen25AudioModel):
@@ -41,8 +41,6 @@ class Qwen4ExpTextModel(_Qwen35MRopeMixin, _LinearAttentionVReorderBase):
        self._ple_shards: dict[int, str] = {}
        self._ple_row_dim: int | None = None
        self._mtp_fc: dict[str, Tensor] = {}
-        # halves of the merged weights, keyed by (bid, merged tensor), see _merge_shared_input
-        self._merge_parts: dict[tuple[int, gguf.MODEL_TENSOR], dict[int, Tensor]] = {}

    @classmethod
    def filter_tensors(cls, item):
@@ -133,79 +131,44 @@ class Qwen4ExpTextModel(_Qwen35MRopeMixin, _LinearAttentionVReorderBase):
        # int64 hash constants must stay exact; 1-D tensors force F32, so use KV
        if name.endswith("ple_embedding.layer_multipliers"):
            self._ple_multipliers = [int(x) for x in data_torch.tolist()]
-            return
+            return []
        if name.endswith("ple_embedding.ngram_heads_offsets"):
            self._ple_head_offsets = [int(x) for x in data_torch.tolist()]
-            return
+            return []
        if name.endswith("ple_embedding.ngram_heads_vocab_sizes"):
            self._ple_head_vocab_sizes = [int(x) for x in data_torch.tolist()]
-            return
+            return []

        if ".ngram_embedding.shard_" in name:
-            yield from self._place_ple_shard(data_torch, name)
-            return
+            return self._place_ple_shard(data_torch, name)

        # eh_proj([e ; h_s]) = fc_embedding(e) + fc_hidden(h_s) for every hc stream s
        if name.endswith((".nextn_fc_embedding.weight", ".nextn_fc_hidden.weight")):
            self._mtp_fc[name.rsplit(".", 2)[1]] = data_torch
            if len(self._mtp_fc) < 2:
-                return
+                return []
            eh = torch.cat([self._mtp_fc.pop("nextn_fc_embedding"), self._mtp_fc.pop("nextn_fc_hidden")], dim=1)
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.NEXTN_EH_PROJ, bid, ".weight"), eh)
-            return
+            return [(self.format_tensor_name(gguf.MODEL_TENSOR.NEXTN_EH_PROJ, bid, ".weight"), eh)]

        # one projection feeds indexer q and k; split it, as minimax-m3 does
        if ".indexer.index_qk_proj.weight" in name:
            n_q = self.hparams["indexer_n_heads"] * self.hparams["indexer_head_dim"]
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.INDEXER_Q_PROJ, bid, ".weight"), data_torch[:n_q])
-            yield (self.format_tensor_name(gguf.MODEL_TENSOR.INDEXER_K_PROJ, bid, ".weight"), data_torch[n_q:])
-            return
+            q = data_torch[:n_q]
+            k = data_torch[n_q:]
+            return [
+                (self.format_tensor_name(gguf.MODEL_TENSOR.INDEXER_Q_PROJ, bid, ".weight"), q),
+                (self.format_tensor_name(gguf.MODEL_TENSOR.INDEXER_K_PROJ, bid, ".weight"), k),
+            ]

        # Gemma zero-centred gammas the inherited norm.weight rule misses
        if name.endswith((".ple.norm_key.weight", ".ple.norm_query.weight", ".ple.norm_conv.weight",
                          ".indexer.q_layernorm.weight", ".indexer.k_layernorm.weight")):
-            yield (self.map_tensor_name(name), data_torch + 1)
-            return
+            return [(self.map_tensor_name(name), data_torch + 1)]

        if name.endswith(".ple.conv1d.weight"):
-            yield (self.map_tensor_name(name), data_torch.squeeze())
-            return
+            return [(self.map_tensor_name(name), data_torch.squeeze())]

-        for new_name, data in super().modify_tensors(data_torch, name, bid):
-            yield from self._merge_shared_input(new_name, data, bid)
-
-    # matmuls that read the same input, written as one weight [a | b] so the runtime does one matmul:
-    # (merged tensor, (a, b), linear-attention layers only)
-    _SHARED_INPUT_MERGES = (
-        (gguf.MODEL_TENSOR.HC_ATTN_DOWN_INJECT, (gguf.MODEL_TENSOR.HC_ATTN_DOWN, gguf.MODEL_TENSOR.HC_ATTN_INJECT), False),
-        (gguf.MODEL_TENSOR.HC_FFN_DOWN_INJECT,  (gguf.MODEL_TENSOR.HC_FFN_DOWN,  gguf.MODEL_TENSOR.HC_FFN_INJECT),  False),
-        (gguf.MODEL_TENSOR.ATTN_QKV_GATE,       (gguf.MODEL_TENSOR.ATTN_QKV,     gguf.MODEL_TENSOR.ATTN_GATE),      True),
-        (gguf.MODEL_TENSOR.SSM_BETA_ALPHA_CAT,  (gguf.MODEL_TENSOR.SSM_BETA,     gguf.MODEL_TENSOR.SSM_ALPHA),      True),
-    )
-
-    # caches one half of a merged weight until the other half arrives, then yields [a | b]
-    def _merge_shared_input(self, new_name: str, data: Tensor, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        # a pre-quantized checkpoint carries per-tensor scales, which one merged weight cannot hold
-        if bid is None or self.hparams.get("quantization_config"):
-            yield (new_name, data)
-            return
-
-        layer_types = self.hparams.get("layer_types") or []
-        is_linear = bid < len(layer_types) and layer_types[bid] == "linear_attention"
-
-        match = next(((merged, parts.index(part)) for merged, parts, linear_only in self._SHARED_INPUT_MERGES
-                      for part in parts
-                      if (is_linear or not linear_only) and self.match_model_tensor_name(new_name, part, bid)), None)
-        if match is None:
-            yield (new_name, data)
-            return
-
-        merged, idx = match
-        halves = self._merge_parts.setdefault((bid, merged), {})
-        halves[idx] = data
-        if len(halves) == 2:
-            del self._merge_parts[(bid, merged)]
-            yield (self.format_tensor_name(merged, bid), torch.cat([halves[0], halves[1]], dim=0))
+        return super().modify_tensors(data_torch, name, bid)

    # the shards concatenate into a tensor of well over 100 GB
    # use LazyChunkedTensor here, a single shard resident at a time
@@ -251,8 +214,6 @@ class Qwen4ExpTextModel(_Qwen35MRopeMixin, _LinearAttentionVReorderBase):
        super().prepare_tensors()
        if self._mtp_fc:
            raise ValueError(f"MTP projection missing its other half: {sorted(self._mtp_fc)}")
-        if self._merge_parts:
-            raise ValueError(f"merged weights missing a half: {sorted((bid, t.name) for bid, t in self._merge_parts)}")
        n_parts = self.hparams.get("split_ngram_parts", 0)
        if self._ple_shards and len(self._ple_shards) != n_parts:
            raise ValueError(
@@ -1,114 +0,0 @@
-from __future__ import annotations
-
-from pathlib import Path
-
-import torch
-
-from .base import MmprojModel, ModelBase, gguf
-from .qwen import Qwen2Model
-
-
-@ModelBase.example("sensenova/SenseNova-U1.5-8B-MoT")
-@ModelBase.register("NEOChatModel")
-class SenseNovaU1Model(Qwen2Model):
-    model_arch = gguf.MODEL_ARCH.SENSENOVA_U1
-
-    @classmethod
-    def filter_tensors(cls, item):
-        name, _ = item
-        if not name.startswith("language_model.") or "_mot_gen" in name:
-            return None
-        return super().filter_tensors(item)
-
-    def set_gguf_parameters(self):
-        if int(self.hparams.get("num_experts") or 0) > 1:
-            raise ValueError("SenseNova U1 conversion currently supports the dense understanding branch")
-        if self.rope_parameters.get("rope_type", "default") not in ("default", None):
-            raise ValueError("SenseNova U1 scaled RoPE is not supported")
-        if self.hparams.get("attention_bias", False):
-            raise ValueError("SenseNova U1 attention bias is not supported")
-        if self.hparams.get("use_sliding_window", False):
-            raise ValueError("SenseNova U1 sliding attention is not supported")
-        super().set_gguf_parameters()
-        self.gguf_writer.add_rope_freq_base_spatial(self.hparams["rope_theta_hw"])
-
-    def get_tensors(self):
-        for name, gen in self.model_tensors.items():
-            if name.endswith((".q_norm_hw.weight", ".k_norm_hw.weight")):
-                continue
-            data = gen()
-            if name.endswith((".q_norm.weight", ".k_norm.weight")):
-                spatial = self.model_tensors[name.replace("_norm.weight", "_norm_hw.weight")]()
-                # Keep one weight vector; the graph normalizes its two halves independently.
-                data = torch.cat((data, spatial), dim=0)
-            yield name, data
-
-
-@ModelBase.register("NEOVisionModel")
-@ModelBase.example("sensenova/SenseNova-U1.5-8B-MoT")
-class SenseNovaU1VisionModel(MmprojModel):
-    def __init__(self, dir_model: Path, *args, **kwargs):
-        hparams = kwargs.pop("hparams", None)
-        if hparams is None:
-            hparams = ModelBase.load_hparams(dir_model, is_mistral_format=False)
-        else:
-            hparams = dict(hparams)
-
-        vision_config = dict(hparams["vision_config"])
-        # NEOVisionModel has no transformer blocks
-        # these defaults satisfy the generic mmproj metadata contract without replacing checkpoint values
-        vision_config.setdefault("image_size", 0)
-        vision_config.setdefault("intermediate_size", 0)
-        vision_config.setdefault("num_hidden_layers", 0)
-        vision_config.setdefault("num_attention_heads", 1)
-        hparams["vision_config"] = vision_config
-        super().__init__(dir_model, *args, hparams=hparams, **kwargs)
-
-        self.preprocessor_config.setdefault("image_mean", [0.485, 0.456, 0.406])
-        self.preprocessor_config.setdefault("image_std", [0.229, 0.224, 0.225])
-
-    @classmethod
-    def filter_tensors(cls, item):
-        name, gen = item
-        if not name.startswith("vision_model."):
-            return None
-        return super().filter_tensors((name, gen))
-
-    def set_gguf_parameters(self):
-        super().set_gguf_parameters()
-        self.gguf_writer.add_clip_projector_type("sensenova_u1")
-        vision_config = self.hparams["vision_config"]
-        # These defaults match the known U1.5 checkpoint when older configs omit fields.
-        hidden_size = int(vision_config.get("hidden_size", 1024))
-        min_pixels = int(vision_config.get("min_pixels", 65536))
-        max_pixels = int(vision_config.get("max_pixels", 16777216))
-        downsample_ratio = float(vision_config.get("downsample_ratio", 0.5))
-        if downsample_ratio <= 0:
-            raise ValueError(f"SenseNova U1 downsample_ratio must be positive, got {downsample_ratio}")
-        merge_size = round(1.0 / downsample_ratio)
-        if abs((1.0 / downsample_ratio) - merge_size) > 1e-6:
-            raise ValueError(f"SenseNova U1 downsample_ratio must have an integer reciprocal, got {downsample_ratio}")
-        norm_eps = float(vision_config.get("layer_norm_eps", vision_config.get("rms_norm_eps", 1e-6)))
-        self.gguf_writer.add_vision_head_dim(hidden_size)
-        self.gguf_writer.add_vision_attention_layernorm_eps(norm_eps)
-        self.gguf_writer.add_vision_min_pixels(min_pixels)
-        self.gguf_writer.add_vision_max_pixels(max_pixels)
-        self.gguf_writer.add_vision_spatial_merge_size(merge_size)
-
-    def modify_tensors(self, data_torch, name, bid):
-        tensor_names = {
-            "vision_model.embeddings.patch_embedding.weight": "v.patch_embd.weight",
-            "vision_model.embeddings.patch_embedding.bias":   "v.patch_embd.bias",
-            "vision_model.embeddings.dense_embedding.weight": self.format_tensor_name(gguf.MODEL_TENSOR.V_MMPROJ, 0),
-            "vision_model.embeddings.dense_embedding.bias": self.format_tensor_name(
-                gguf.MODEL_TENSOR.V_MMPROJ, 0, suffix=".bias"
-            ),
-        }
-        if name not in tensor_names:
-            raise ValueError(f"Unexpected SenseNova U1 vision tensor: {name!r}")
-        return [(tensor_names[name], data_torch)]
-
-    def tensor_force_quant(self, name, new_name, bid, n_dims):
-        if new_name == self.format_tensor_name(gguf.MODEL_TENSOR.V_MMPROJ, 0):
-            return gguf.GGMLQuantizationType.F16 if self.ftype == gguf.LlamaFileType.MOSTLY_F16 else gguf.GGMLQuantizationType.F32
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
@@ -1,116 +0,0 @@
-from __future__ import annotations
-
-import json
-from pathlib import Path
-from typing import Any
-
-import torch
-
-from .base import ModelBase, MmprojModel, gguf, logger
-from .qwen import Qwen3Model
-
-
-def _is_soprano_checkpoint(dir_model: Path) -> bool:
-    if not (dir_model / "decoder.pth").is_file() or not (dir_model / "config.json").is_file():
-        return False
-    with open(dir_model / "config.json", encoding="utf-8") as f:
-        config = json.load(f)
-    if config.get("model_type") != "qwen3" or config.get("bos_token_id") != 3 or config.get("eos_token_id") != 3:
-        return False
-    if not (dir_model / "tokenizer.json").is_file():
-        return False
-    with open(dir_model / "tokenizer.json", encoding="utf-8") as f:
-        added = {t["id"]: t["content"] for t in json.load(f).get("added_tokens", [])}
-    return [added.get(i) for i in range(4)] == ["[UNK]", "[TEXT]", "[START]", "[STOP]"]
-
-
-@ModelBase.register_hparams_loader(_is_soprano_checkpoint)
-def _load_soprano_hparams(dir_model: Path) -> dict[str, Any]:
-    logger.info("gguf: detected Soprano checkpoint")
-    hparams = ModelBase.load_hparams(dir_model, False, guess=False)
-    # the text model stays Qwen3, the mmproj holds the decoder
-    hparams["text_config"] = {"architectures": ["SopranoTextModel"]}
-    hparams["architectures"] = ["SopranoModel"]
-    return hparams
-
-
-@ModelBase.register("SopranoTextModel")
-class SopranoTextModel(Qwen3Model):
-    model_arch = gguf.MODEL_ARCH.QWEN3
-
-    def get_vocab_base_pre(self, tokenizer) -> str:
-        return "soprano"
-
-    def set_vocab(self):
-        super().set_vocab()
-        self.gguf_writer.add_unk_token_id(0)
-
-
-@ModelBase.register("SopranoModel")
-@ModelBase.example("ekwek/Soprano-80M", "ekwek/Soprano-1.1-80M")
-class SopranoModel(MmprojModel):
-    has_vision_encoder = False
-    has_audio_encoder = False
-
-    def _load_decoder(self) -> dict[str, torch.Tensor]:
-        state = torch.load(self.dir_model / "decoder.pth", map_location="cpu", mmap=True, weights_only=True)
-        return state.get("state_dict", state)
-
-    def get_audio_config(self):
-        state = self._load_decoder()
-        n_embd, n_inp, _ = state["decoder.embed.weight"].shape
-        n_ff = state["decoder.convnext.0.pwconv1.weight"].shape[0]
-        n_layer = sum(1 for name in state if name.endswith(".pwconv1.weight"))
-        if n_inp != self.n_embd_text:
-            raise ValueError(f"Soprano decoder input dim {n_inp} does not match the backbone dim {self.n_embd_text}")
-        return {"num_hidden_layers": n_layer, "hidden_size": n_embd, "intermediate_size": n_ff}
-
-    def set_gguf_parameters(self):
-        self.gguf_writer.add_file_type(self.ftype)
-        self.gguf_writer.add_clip_has_gen_audio_encoder(True)
-        self.gguf_writer.add_clip_gen_audio_projector_type(gguf.VisionProjectorType.SOPRANO)
-        self.gguf_writer.add_gen_audio_projection_dim(self.n_embd_text)
-        self.gguf_writer.add_gen_audio_embedding_length(self.hparams["hidden_size"])
-        self.gguf_writer.add_gen_audio_feed_forward_length(self.hparams["intermediate_size"])
-        self.gguf_writer.add_gen_audio_block_count(self.hparams["num_hidden_layers"])
-        self.gguf_writer.add_gen_audio_head_count(1)
-        self.gguf_writer.add_gen_audio_attention_layernorm_eps(1e-6)
-
-    def get_tensors(self):
-        yield from self._load_decoder().items()
-
-    def tensor_force_quant(self, name, new_name, bid, n_dims):
-        if "dwconv.weight" in new_name or (name == "decoder.embed.weight" and n_dims == 3):
-            return gguf.GGMLQuantizationType.F16
-        return super().tensor_force_quant(name, new_name, bid, n_dims)
-
-    def modify_tensors(self, data_torch, name, bid):
-        if name == "head.istft.window":
-            # the ISTFT in mtmd builds its own window
-            if not torch.equal(data_torch.float(), torch.hann_window(data_torch.shape[0])):
-                raise ValueError("Soprano requires a periodic Hann window")
-            return
-        if name == "decoder.embed.weight" and data_torch.shape[-1] == 1:
-            data_torch = data_torch.squeeze(-1)
-
-        T = gguf.MODEL_TENSOR
-        if name.startswith("decoder.convnext."):
-            _, _, blk, key, *suffix = name.split(".")
-            block_map = {
-                "dwconv":  T.A_GEN_WAV_UP_DWCONV,
-                "norm":    T.A_GEN_WAV_UP_NORM,
-                "pwconv1": T.A_GEN_WAV_UP_PW1,
-                "pwconv2": T.A_GEN_WAV_UP_PW2,
-                "gamma":   T.A_GEN_WAV_UP_GAMMA,
-            }
-            yield self.format_tensor_name(block_map[key], int(blk), suffix="".join("." + s for s in suffix)), data_torch
-            return
-
-        tensor_map = {
-            "decoder.embed":            T.A_GEN_WAV_INPUT,
-            "decoder.norm":             T.A_GEN_WAV_NORM,
-            "decoder.final_layer_norm": T.A_GEN_WAV_OUTPUT_NORM,
-            "head.out":                 T.A_GEN_WAV_OUTPUT,
-        }
-        key, suffix = name.rsplit(".", 1)
-        yield self.format_tensor_name(tensor_map[key], suffix="." + suffix), data_torch
@@ -58,7 +58,7 @@ def parse_args() -> argparse.Namespace:
    )
    parser.add_argument(
        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0", "tq1_0", "tq2_0", "auto"], default="auto",
-        help="output format - use f32 for float32, f16 for float16, bf16 for bfloat16, q8_0 for Q8_0, tq1_0 or tq2_0 for ternary, and auto to preserve supported FP8 weights and otherwise select the highest-fidelity 16-bit float type",
+        help="output format - use f32 for float32, f16 for float16, bf16 for bfloat16, q8_0 for Q8_0, tq1_0 or tq2_0 for ternary, and auto for the highest-fidelity 16-bit float type",
    )
    parser.add_argument(
        "--bigendian", action="store_true",
@@ -127,7 +127,7 @@ def parse_args() -> argparse.Namespace:
    )
    parser.add_argument(
        "--dspark", action="store_true",
-        help="Export only the DeepSeek-V4 or Mistral Large 4 DSpark draft tensors as a separate GGUF.",
+        help="Export only the DeepSeek-V4 DSpark draft tensors as a separate GGUF.",
    )
    parser.add_argument(
        "--mistral-format", action="store_true",
@@ -255,9 +255,6 @@ def main() -> None:
            assert hparams.get("vision_encoder") is not None, "This model does not support multimodal"
            from conversion.pixtral import PixtralModel
            model_class = PixtralModel
-        elif hparams.get("moe") is not None and hparams.get("dsa") is not None:
-            from conversion.mistral4 import MistralLarge4Model
-            model_class = MistralLarge4Model
        elif hparams.get("moe") is not None:
            from conversion.mistral import MistralMoeModel
            model_class = MistralMoeModel
@@ -270,21 +267,11 @@ def main() -> None:
            sys.exit(1)

        if args.dspark:
-            if is_mistral_format or model_architecture == "MistralLarge4ForConditionalGeneration":
-                if model_class.__name__ != "MistralLarge4Model":
-                    logger.error("--dspark is only supported for Mistral Large 4 among Mistral-format models")
-                    sys.exit(1)
-                from conversion.mistral4 import MistralLarge4DSparkModel
-                model_class = MistralLarge4DSparkModel
-            elif model_architecture not in ("DeepseekV4ForCausalLM", "DeepseekV41ForCausalLM"):
-                logger.error("--dspark is only supported for DeepseekV4ForCausalLM, DeepseekV41ForCausalLM and Mistral Large 4")
+            if is_mistral_format or model_architecture != "DeepseekV4ForCausalLM":
+                logger.error("--dspark is only supported for DeepseekV4ForCausalLM")
                sys.exit(1)
-            elif model_architecture == "DeepseekV41ForCausalLM":
-                from conversion.deepseek41 import DeepseekV41DSparkModel
-                model_class = DeepseekV41DSparkModel
-            else:
-                from conversion.deepseek import DeepseekV4DSparkModel
-                model_class = DeepseekV4DSparkModel
+            from conversion.deepseek import DeepseekV4DSparkModel
+            model_class = DeepseekV4DSparkModel

        if args.mtp or args.no_mtp:
            if not model_class.supports_mtp_export:
@@ -101,7 +101,6 @@ models = [
    {"name": "command-r",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", },
    {"name": "tiny_aya",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereLabs/tiny-aya-base", },
    {"name": "cohere2moe",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", },
-    {"name": "flint",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/RecursiveSI/flint", },
    {"name": "qwen2",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", },
    {"name": "olmo",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", },
    {"name": "dbrx",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", },
@@ -178,7 +177,6 @@ pre_computed_hashes = [
    {"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/zai-org/GLM-4.5-Air", "chkhsh": "9ca2dd618e8afaf09731a7cf6e2105b373ba6a1821559f258b272fe83e6eb902"},
    {"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/zai-org/GLM-4.7-Flash", "chkhsh": "cdf5f35325780597efd76153d4d1c16778f766173908894c04afc20108536267"},
    {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", "chkhsh": "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35"},
-    {"name": "domyn-small", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/domyn/Domyn-Small-v1.0", "chkhsh": "6e4d72c6e5260b90606e26a0931fa15e4b42ec6e97719d12919f2b1f93478262"},
    {"name": "hunyuan", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "chkhsh": "7e57df22b1fe23a7b1e1c7f3dc4e3f96d43a4eb0836d0c6bdc3436d7b2f1c664"},
    {"name": "hunyuan-dense", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-4B-Instruct", "chkhsh": "bba3b3366b646dbdded5dbc42d59598b849371afc42f7beafa914afaa5b70aa6"},
    {"name": "hy_v4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hy4-preview", "chkhsh": "e6ddf9c6686791c12d698d34c31ab9be1fea9af5a3d9a6909783ab382198ae1c"},
@@ -72,8 +72,6 @@ However, A6x GPUs in phones are likely not supported due to the outdated driver
 | MXFP4                  | Support                    |
 | IQ4_NL                 | Support                    |

-`MUL_MAT_ID` accepts an expert id of `-1` and writes zeros to that slot. `ADD_ID` adds no bias for `-1`, including in fused MoE operations.
-
 ## Model Preparation

 Since common quantizations are supported now, it is recommanded to download GGUF models directly from Huggingface.
@@ -783,8 +783,7 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | GGML_SYCL          | ON (mandatory)                        | Enable build with SYCL code path.           |
 | GGML_SYCL_TARGET   | INTEL *(default)*                     | Set the SYCL target device type.            |
 | GGML_SYCL_DEVICE_ARCH | Optional                           | Set the SYCL device architecture. Setting the device architecture can improve the performance. See the table [--offload-arch](https://github.com/intel/llvm/blob/sycl/sycl/doc/design/OffloadDesign.md#--offload-arch) for a list of valid architectures. |
-| GGML_SYCL_F16      | OFF *(default)* \|ON *(optional)*     | Enable FP16 build with SYCL code path. [1.](#sycl-f16-footnote) |
-| GGML_SYCL_SUPPORT_XMX | ON *(default)* \|OFF *(Optional)* | Build the code paths that use XMX (DPAS) instructions. Set OFF for AOT builds targeting iGPUs without XMX, which otherwise fail to compile. |
+| GGML_SYCL_F16      | OFF *(default)* \|ON *(optional)*     | Enable FP16 build with SYCL code path. (1.) |
 | GGML_SYCL_GRAPH    | ON *(default)* \|OFF *(Optional)*     | Enable build with [SYCL Graph extension](https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/experimental/sycl_ext_oneapi_graph.asciidoc). |
 | GGML_SYCL_DNN      | ON *(default)* \|OFF *(Optional)*     | Enable build with oneDNN.                   |
 | GGML_SYCL_HOST_MEM_FALLBACK | ON *(default)* \|OFF *(Optional)* | Allow host memory fallback when device memory is full during quantized weight reorder. Enables inference to continue at reduced speed (reading over PCIe) instead of failing. Requires Linux kernel 6.8+. |
@@ -792,7 +791,7 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | CMAKE_C_COMPILER   | `icx` *(Linux)*, `icx/cl` *(Windows)* | Set `icx` compiler for SYCL code path.      |
 | CMAKE_CXX_COMPILER | `icpx` *(Linux)*, `icx` *(Windows)*   | Set `icpx/icx` compiler for SYCL code path. |

-1. <a id="sycl-f16-footnote"></a>FP32 or FP16 have different performance impact to LLM. Recommended to test them for better prompt processing performance on your models. You need to rebuild the code after change `GGML_SYCL_F16=OFF/ON`.
+1. FP32 or FP16 have different performance impact to LLM. Recommended to test them for better prompt processing performance on your models. You need to rebuild the code after change `GGML_SYCL_F16=OFF/ON`.

 ### Runtime

@@ -803,8 +802,7 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | GGML_SYCL_DEV2DEV_MEMCPY | 0 (default), 1, 2 | Choose the method of dev2dev memory copy.<br>Value: <br>*  0: SYCL API (default), only support dGPUs.<br>* 1: L0 API -- Better performance, only support dGPUs, found to lead to abnormal crash in some case. <br>* 2: Host Forward -- Most stable method for all cases (including iGPU + dGPU*N), but with lower performance (-2% to -5%).<br>SYCL & L0 API are easy to be impacted by Intel GPU driver issue. When you meet the garbled output or crash issues in multiple GPUs case, try with this debug flag to work around or check the issue.|
 | GGML_SYCL_ENABLE_FLASH_ATTN | 1 (default) or 0| Enable Flash-Attention. It can reduce memory usage. The performance impact depends on the LLM.|
 | GGML_SYCL_ENABLE_OPT | 0 or 1 (default)| Enable optimize features for Intel GPUs. (Recommended to 0 for Intel devices older than Gen 10) |
-| GGML_SYCL_ENABLE_GRAPH | 0 (default) or 1 | Enable running computations through SYCL Graphs feature. Disabled by default because SYCL Graph is still in development, performance depends on model and device. [1.](#sycl-graph-footnote) |
-| GGML_SYCL_GRAPH_EVICTION_TIMEOUT | 10 (default) or positive integer | Seconds a cached SYCL graph can go unused before it is evicted. Checked on a sweep that runs every half of this timeout. |
+| GGML_SYCL_ENABLE_GRAPH | 0 (default) or 1 | Enable running computations through SYCL Graphs feature. Disabled by default because SYCL Graph is still on development, no better performance. |
 | GGML_SYCL_ENABLE_HOST_PINNED_MEM | 0 or 1 (default) | Enable host pinned memory to speed up copy data from host to device. When disable it, host memory will common malloc() on CPU. Disable it when use `--load-model mlock`.|
 | GGML_SYCL_HOST_PINNED_MEM_2G | 0 (default) or 1 | Limit the max memory allocation to be no more than 2GB when enable host pinned memory. USM allocations above 2 GiB take the relaxed/large-allocation path, which serializes H2D copies with compute and prevents copy/compute overlap. It will impact the startup time. Need more test. Depend on `GGML_SYCL_ENABLE_HOST_PINNED_MEM=1`.|
 | GGML_SYCL_UPLOAD_STAGING_SLOTS | 4 (default) or non-negative integer | Number of 8 MiB pinned host slots used to stage tensor uploads (model loading), so the host copy of one slot overlaps the transfer of the previous one. Set to 0 to use the old path: a malloc'd bounce buffer and a blocking copy per tensor. |
@@ -813,7 +811,6 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | GGML_SYCL_ENABLE_DNN | 0 or 1 (default)| Enable running computations through oneDNN and always use oneMKL. |
 | GGML_SYCL_FA_ONEDNN | 1 (default) or 0 | Enable the oneDNN fused SDPA (flash-attention) path on supported GPUs. Set to 0 to always use the native SYCL flash-attention kernel. |
 | GGML_SYCL_FA_ONEDNN_MAX_KV | 0 (default, disabled) or positive integer | By default (0), all sequences are handled by the oneDNN fused SDPA path, regardless of KV length; a positive value caps that length, past which sequences fall back to the native kernel. If GPU driver watchdog resets (DEVICE_LOST) occur during long-context inference, set this near the context depth where they start, e.g. 24576. |
-| GGML_SYCL_FA_DPAS | 1 (default) or 0 | Enable the XMX flash-attention kernel for a q4_0 or q8_0 KV cache (`-ctk q4_0 -ctv q4_0` or `-ctk q8_0 -ctv q8_0`) on Xe2 or newer GPUs (Arc B-series, Lunar Lake, Panther Lake, Wildcat Lake). It reads the quantized cache directly for decode and batches of up to 31 tokens, head sizes 64, 96, 128, 256 and 512, up to 32 query heads per KV head. Set to 0 to use the native SYCL flash-attention kernels. |
 | GGML_SYCL_ENABLE_VMM | 0 or 1 (default) | Enable the virtual-memory device pool. |
 | GGML_SYCL_ENABLE_MKL_FA | 1 (default) or 0 | Enable oneMKL GEMM flash attention for XMX-accelerated prompt processing with quantized KV cache. Automatically activates during prefill (prompt processing) when all conditions are met: (1) flash-attn enabled (`-fa` or `--flash-attn on`), (2) KV cache quantized (`--cache-type-k q8_0 --cache-type-v q8_0` or other `*_0/*_1` types), (3) batch size ≥ 1024 (`--batch-size 1024`), (4) prompt length ≥ 1024 tokens. Set to 0 to force the TILE kernel for A/B testing. Example minimum command: `llama-cli -m model.gguf -fa -ngl 99 --cache-type-k q8_0 --cache-type-v q8_0 --batch-size 1024 -p "your prompt"` |
 | GGML_SYCL_MKL_FA_DEBUG | 0 (default) or 1 | Enable per-call diagnostic logging for MKL flash attention: GEMM/softmax timings, interleaved-head detection, and buffer memory usage. |
@@ -822,9 +819,6 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | GGML_SYCL_MKL_FA_DIAG | 0 (default) or 1 | Enable output fingerprinting for MKL flash attention. Dumps the first 64 float output values for the first 6 FA calls with n_kv ≥ 1024, labeled with kernel type (MKL/TILE/VEC) for cross-kernel comparison. |
 | GGML_SYCL_ENABLE_FUSION | 0 or 1 (default) | Enable fused-kernel dispatch in graph compute. Unsupported types and layouts fall back to the standalone op kernels. See `ggml_sycl_can_fuse()`. |
 | GGML_SYCL_ENABLE_ESIMD | 0 or 1 (default)| Enable ESIMD kernels when available. |
-| GGML_SYCL_ENABLE_XMX_ESIMD | 0 or 1 (default) | Use the XMX (DPAS) multi-column mat-vec kernels for reordered Q2_K to Q6_K and Q8_0 weights on Xe2 GPUs. Speeds up speculative decoding and short prompt batches. |
-| GGML_SYCL_XMX_MAX_COLS | non-negative integer, 80 (default) | Most activation columns (batch size) the `GGML_SYCL_ENABLE_XMX_ESIMD` kernels take; wider batches use the other mat-mul kernels. A larger value keeps XMX for bigger batches (as repeated launches), `0` never uses it. |
-| GGML_SYCL_XMX_GLU_MAX_COLS | 0 to 16, 16 (default) | Most columns the fused gate and up XMX kernel takes; wider batches run gate and up as separate mat-vecs. 16 is the most the kernel is built for, larger values are clamped to 16 and `0` turns the fused kernel off. |
 | GGML_SYCL_XMX_GATHER_TYPES | decimal bitmask, all bits set (default) | Weight formats that may use the XMX dequant-GEMM paths, which dequantize weights straight into the XMX tiles. This speeds up prompt processing of MoE models on GPUs with XMX units (Arc A- and B-series, Arc Pro, Data Center GPU Max), for example pp512 of Qwen3-30B-A3B UD-IQ3_XXS by about 50% on an Arc Pro B60. Bits:<br>* 1: IQ4_NL, 2: IQ3_S, 4: IQ4_XS, 8: IQ3_XXS, 16: IQ2_XXS, 32: IQ2_XS, 64: IQ2_S, 128: IQ1_S, 256: IQ1_M<br>* 512: Q8_0, 1024: Q4_K, 2048: Q5_K, 4096: Q6_K (MoE `MUL_MAT_ID` only)<br>Add values to combine them, for example `3` for IQ4_NL and IQ3_S; `0` disables the paths. A set bit does not force the path: batches of more than 64 tokens per expert or row lengths that are not a multiple of 256 (32 for IQ4_NL and Q8_0) use the library GEMM. |
 | GGML_SYCL_XMX_GATHER_SHAPES | decimal bitmask, 255 (default) | XMX `joint_matrix` combinations the paths of `GGML_SYCL_XMX_GATHER_TYPES` may use; the operand type comes from `GGML_SYCL_DYNAMIC_PRECISION` and the best supported combination is picked automatically (logged as `fg_pick_combo`). Bits:<br>* Xe2, Xe3, Xe-HPC: 1: f16 8x16x16, 2: f16 16x16x16, 4: f16 32x64x16, 8: f16 32x64x32, 32: tf32 8x16x8, 64: bf16 8x16x16<br>* Xe-HPG (Arc A770, ARL-H): 16: f16 8x8x16, 128: bf16 8x8x16<br>Clear a bit to exclude a combination, or set a single bit to force one for testing. |
 | GGML_SYCL_DYNAMIC_PRECISION | `F16` (default with `GGML_SYCL_F16=ON`), `BF16`, `TF32` or `F32` (default otherwise) | Operand type of the XMX dequant-GEMM paths (`GGML_SYCL_XMX_GATHER_TYPES`); accumulation is always f32. `F16` is the fastest, but activations above 65504 overflow. `BF16` keeps the f32 range at a 7-bit mantissa, `TF32` keeps the range and the f16 mantissa but is about 30% slower and needs Xe2, Xe3 or Xe-HPC, and `F32` turns the XMX paths off. Ops that request a higher src1 precision ([TAG_GGML_PREC]) get it regardless of this setting. |
@@ -837,29 +831,6 @@ User can use the device management in [docs/multi-gpu.md](https://github.com/ggm
 | UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS | 0 (default) or 1 | Allow SYCL/Unified Runtime Level Zero device allocations larger than 4 GiB. llama.cpp's direct Level Zero allocation path requests the relaxed maximum-size limit itself when GGML_SYCL_ENABLE_LEVEL_ZERO=1. |
 | GGML_SYCL_USM_SYSTEM | 0 (default) or 1 | Enable experimental support for [USM system allocations](https://github.khronos.org/SYCL_Reference/iface/usm_basic_concept.html#system-allocations) for large GPU buffers. This requires enough host memory for model weights and caches, an Intel Xe2+ GPU such as BMG or newer and supported on Linux only, with CONFIG_DRM_XE_GPUSVM enabled. |

-1. <a id="sycl-graph-footnote"></a>SYCL Graph support (`GGML_SYCL_ENABLE_GRAPH=1`) is limited to single-GPU workloads and may also depend on the model architecture and kernels in use.
-   Performance depends on how often a recorded graph can be reused, which will vary by workload.
-   For this reason, it is recommended to measure performance both with and without graphs enabled on your workloads.
-
-   To check whether graphs are supported for your workload, enable INFO logging (`-lv 4`) and look for `enabling SYCL graphs` or `disabling SYCL graphs`.
-   Support may differ between prefill and decode, so check the status for the phases that matter to you.
-   DEBUG logging (`-lv 5`) reports why graphs are disabled.
-
-   Then benchmark with your preferred tool, such as `llama-bench`.
-   For graph activity details (including reuse), set `GGML_SYCL_DEBUG=1` and look for `[SYCL-GRAPH]` messages.
-
-### Flash-attention kernels
-
-Attention dispatches on shape and hardware.
-
-DPAS is the quantized path. q4_0 or q8_0 KV cache, decode and batches up to 31 tokens, Xe2 and newer only. It issues dpas straight on the quantized blocks and never materialises an f16 copy. vISA gets used here because joint_matrix has no s4 type.
-
-oneDNN does XMX through SDPA but converts KV to f16 first, so it ends up on prefill: 32+ tokens and 1K+ keys. MKL is the prompt processing equivalent through oneMKL GEMM, also f16.
-
-VEC/TILE are plain kernels with no XMX in them. They run when DPAS is off or the hardware doesn't support it.
-
-joint_matrix belongs to the matmul GEMMs. Attention doesn't use it.
-
 ## Compile-time Flags

 Pass these via `CXXFLAGS` or add a one-off `#define` to enable a flag on the spot.
@@ -324,9 +324,6 @@ on 4 physical NPUs, or `--devices 'HTP0[0-1:0],HTP1[0-1:1]'` on 2 physical NPUs
 - `GGML_HEXAGON_NHVX=0`
  Controls the number of HVX hardware threads to use. The default is all (actual number varies depending on the hardware version).

- `GGML_HEXAGON_VTCM_SIZE=0`
-  Overrides the VTCM size used by each session. Accepts `K`, `M` and `G` suffixes; a plain number is in MB. The default (0) uses the size the hardware reports. Values above the hardware size will fail to start the session.
-
 - `GGML_HEXAGON_HOSTBUF=1` (default: 0, disabled)
  Enables allocating host buffers for debugging. By default, host buffers are disabled.

@@ -14,22 +14,22 @@ Legend:

 | Operation | BLAS | CANN | CPU | CUDA | ET | HTP | MTL | OpenCL | SYCL | Vulkan | WebGPU | ZenDNN | zDNN |
 |-----------|------|------|------|------|------|------|------|------|------|------|------|------|------|
-|                              ABS | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              ABS | ❌ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                              ACC | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                              ADD | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              ADD | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             ADD1 | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                           ADD_ID | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                           ARANGE | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                           ARGMAX | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                           ARGMAX | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                          ARGSORT | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                             CEIL | ❌ | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                             CEIL | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                            CLAMP | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                        COL2IM_1D | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                           CONCAT | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ | ❌ |
 |                             CONT | ❌ | 🟡 | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ | ❌ |
-|                       CONV_1D_DW | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ |
+|                       CONV_1D_DW | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
 |                          CONV_2D | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ❌ | ❌ |
-|                       CONV_2D_DW | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                       CONV_2D_DW | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                          CONV_3D | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                CONV_TRANSPOSE_1D | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                CONV_TRANSPOSE_2D | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
@@ -38,97 +38,96 @@ Legend:
 |                              CPY | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
 |               CROSS_ENTROPY_LOSS | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |          CROSS_ENTROPY_LOSS_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                           CUMSUM | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                             DIAG | ❌ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                           CUMSUM | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                             DIAG | ❌ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                    DIAG_MASK_INF | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                              DIV | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              DIV | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                     DSV4_HC_COMB | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                     DSV4_HC_POST | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                      DSV4_HC_PRE | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                              DUP | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                              ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                              EXP | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                            EXPM1 | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                             FILL | ❌ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              DUP | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | 🟡 | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
+|                              ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              EXP | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            EXPM1 | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                             FILL | ❌ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                   FLASH_ATTN_EXT | ❌ | 🟡 | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
-|                            FLOOR | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                  GATED_DELTA_NET | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            FLOOR | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                  GATED_DELTA_NET | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | 🟡 | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                GATED_LINEAR_ATTN | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                            GEGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                        GEGLU_ERF | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                      GEGLU_QUICK | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            GEGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                        GEGLU_ERF | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                      GEGLU_QUICK | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             GELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                         GELU_ERF | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                         GELU_ERF | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                       GELU_QUICK | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                         GET_ROWS | ❌ | 🟡 | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ❌ | ❌ |
 |                    GET_ROWS_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ |
 |                       GROUP_NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                      HARDSIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                        HARDSWISH | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                      HARDSIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                        HARDSWISH | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                           IM2COL | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                        IM2COL_3D | ❌ | ❌ | ✅ | ✅ | ❌ | 🟡 | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                          L2_NORM | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ | ❌ |
+|                        IM2COL_3D | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
+|                          L2_NORM | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ❌ | ✅ | ✅ | 🟡 | ❌ | ❌ |
 |                       LEAKY_RELU | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 | 🟡 | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                LIGHTNING_INDEXER | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ | ❌ |
+|                LIGHTNING_INDEXER | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                              LOG | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             MEAN | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | 🟡 | ✅ | ❌ | ❌ | ❌ |
-|                              MUL | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              MUL | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                          MUL_MAT | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 |
-|                 MUL_MAT_HADAMARD | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | 🟡 | ✅ | ✅ | ✅ | ❌ | 🟡 |
+|                 MUL_MAT_HADAMARD | ❌ | ❌ | ✅ | ❌ | ✅ | 🟡 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 |
 |                       MUL_MAT_ID | ❌ | 🟡 | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | 🟡 | ❌ |
-|                  MUL_MAT_ID_W4A4 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                  MUL_MAT_ID_W4A8 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                     MUL_MAT_W4A4 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                     MUL_MAT_W4A8 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                              NEG | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                  MUL_MAT_ID_W4A4 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
+|                  MUL_MAT_ID_W4A8 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
+|                     MUL_MAT_W4A4 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
+|                     MUL_MAT_W4A8 | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
+|                              NEG | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             NORM | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | 🟡 | ❌ | ❌ |
 |                   OPT_STEP_ADAMW | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                     OPT_STEP_SGD | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                         OUT_PROD | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ |
 |                              PAD | ❌ | 🟡 | ✅ | 🟡 | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ❌ | ❌ |
 |                   PAD_REFLECT_1D | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
-|                          POOL_1D | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                          POOL_2D | ❌ | 🟡 | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                            REGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                          POOL_1D | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
+|                          POOL_2D | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
+|                            REGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             RELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                           REPEAT | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ | ❌ |
 |                      REPEAT_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                         RMS_NORM | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                         RMS_NORM | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                    RMS_NORM_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                             ROLL | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                             ROPE | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                        ROPE_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                            ROUND | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            ROUND | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                        RWKV_WKV6 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                        RWKV_WKV7 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                            SCALE | ❌ | 🟡 | ✅ | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                              SET | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | ❌ | ❌ |
+|                            SCALE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              SET | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ |
 |                         SET_ROWS | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
-|                              SGN | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              SGN | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                          SIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             SILU | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                        SILU_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
 |                              SIN | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                         SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                         SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                         SOFT_MAX | ❌ | 🟡 | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                    SOFT_MAX_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ |
-|                        SOLVE_TRI | ❌ | ❌ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | ❌ | ❌ |
+|                        SOLVE_TRI | ❌ | ❌ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ |
 |                              SQR | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                             SQRT | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                         SSM_CONV | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                         SSM_SCAN | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ❌ |
-|                             STEP | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                              SUB | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                         SSM_SCAN | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | 🟡 | 🟡 | ✅ | ❌ | ❌ |
+|                             STEP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                              SUB | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                              SUM | ❌ | 🟡 | ✅ | 🟡 | ❌ | 🟡 | 🟡 | ❌ | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
 |                         SUM_ROWS | ❌ | ✅ | ✅ | 🟡 | ❌ | 🟡 | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ | ❌ |
-|                           SWIGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                     SWIGLU_CLAMP | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
+|                           SWIGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                     SWIGLU_CLAMP | ❌ | ❌ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                       SWIGLU_OAI | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                             TANH | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                             TANH | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |               TIMESTEP_EMBEDDING | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|                            TOP_K | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | 🟡 | ✅ | ❌ | ❌ |
-|                       TOP_K_MASK | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
-|                              TRI | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
-|                            TRUNC | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            TOP_K | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
+|                              TRI | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
+|                            TRUNC | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                          UPSCALE | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ | ✅ | 🟡 | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                            XIELU | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
@@ -100,9 +100,8 @@ llama-server -m Qwen3-4B.gguf -md Qwen3-4B-DSpark.gguf \

 `--spec-draft-n-max` is clamped to the draft model's trained block size.

-`--spec-draft-p-min P` truncates each drafted block at the first position whose predicted
-acceptance (from the draft's confidence head) falls below `P` (default 0 = disabled).
-Values above 0 require a draft with a confidence head.
+`--spec-draft-conf-min P` truncates each drafted block at the first position whose predicted
+acceptance (from the draft's confidence head, if present) falls below `P` (default 0 = disabled).

 Currently only drafts with a Qwen3 backbone are supported; support for other backbones
 (e.g. Gemma4) is planned.
@@ -1,211 +0,0 @@
-#!/usr/bin/env python3
-"""Example client for the llama-server /v1/realtime websocket, with the openai SDK.
-
-Start a server with an audio model, e.g. LFM3-Audio:
-
-    llama-server -hf LiquidAI/LFM3-Audio-2B-GGUF
-
-Spoken conversation, the audio of each wav file is one user turn. It is sent in chunks as it would come from a
-microphone, the server prefills it as it arrives so the response starts right after the last chunk:
-
-    python examples/realtime_client.py --instructions "Respond with interleaved text and audio." question.wav
-
-Output modalities: left out, the model picks; "text", "audio" or "text,audio" (interleaved) force the response.
-Without --instructions, a forced mode comes with the system prompt of the LFM3-Audio or LFM3-Omni model card:
-
-    python examples/realtime_client.py --output-modalities text,audio --out answer.wav question.wav
-
-Tool calls: --tools gives the conversation a demo tool, the client runs its calls and sends the results back:
-
-    python examples/realtime_client.py --tools --text "Set the volume to 30."
-
-Text turns go in with --text, transcription sessions transcribe each file:
-
-    python examples/realtime_client.py --text "What is the capital of France?"
-    python examples/realtime_client.py --type transcription --language en speech.wav
-
-Live transcription models (Voxtral Realtime, LFM3-Audio live ASR) transcribe while the audio is sent, a transcription
-session opens with ?intent=transcription:
-
-    python examples/realtime_client.py --type transcription speech.wav
-
-Requirements: pip install "openai[realtime]" numpy
-"""
-
-import argparse
-import asyncio
-import base64
-import json
-import time
-import wave
-
-import numpy as np
-from openai import AsyncOpenAI
-
-
-# system prompts of the LiquidAI/LFM3-Audio-2B and LiquidAI/LFM3-Omni-2B model cards for a forced output mode
-MODE_INSTRUCTIONS = {
-    "audio": {"text,audio": "Respond with interleaved text and audio."},
-    "omni":  {"text": "Respond in text.", "text,audio": "Respond in interleaved text and audio."},
-}
-# and with tools: calls, or an answer in interleaved text and speech
-TOOLS_INSTRUCTIONS = "Respond with function calls, and interleaved text and audio.\n\nList of tools:"
-
-# the demo tool of --tools
-DEMO_TOOLS = [{
-    "type": "function",
-    "name": "set_volume",
-    "description": "Set the device speaker volume.",
-    "parameters": {
-        "type": "object",
-        "properties": {"volume": {"type": "integer", "description": "Volume from 0 to 100."}},
-        "required": ["volume"],
-    },
-}]
-
-
-def run_tool(name, arguments):
-    if name == "set_volume":
-        return {"volume": json.loads(arguments)["volume"]}
-    return {"error": f"unknown tool {name}"}
-
-
-def read_wav(path, rate):
-    # PCM16 wav -> mono int16 at the given rate
-    with wave.open(path, "rb") as f:
-        assert f.getsampwidth() == 2, "the example reads PCM16 wav files"
-        x = np.frombuffer(f.readframes(f.getnframes()), dtype=np.int16).astype(np.float32)
-        x = x.reshape(-1, f.getnchannels()).mean(axis=1)
-        sr = f.getframerate()
-    if sr != rate:
-        n = int(round(len(x) * rate / sr))
-        x = np.interp(np.arange(n) * sr / rate, np.arange(len(x)), x)
-    return np.clip(x, -32768, 32767).astype(np.int16)
-
-
-def write_wav(path, pcm, rate):
-    with wave.open(path, "wb") as f:
-        f.setnchannels(1)
-        f.setsampwidth(2)
-        f.setframerate(rate)
-        f.writeframes(pcm.tobytes())
-
-
-async def main(args):
-    # the SDK derives a wss:// URL from base_url, a plain http server needs ws://
-    client = AsyncOpenAI(base_url=args.url, api_key=args.api_key, websocket_base_url=args.url.replace("http", "ws", 1))
-    query = {"intent": "transcription"} if args.type == "transcription" else {}
-    async with client.realtime.connect(model=args.model, extra_query=query) as conn:
-        ev = await conn.recv()
-        if ev.type == "error":
-            print("error:", ev.error.message)
-            return
-        # the audio goes in at the rate of the model, PCM16 mono
-        session_info = ev.model_dump(warnings=False)["session"]
-        audio_fmt = session_info["audio"]
-        card = "omni" if "omni" in str(session_info.get("model", "")).lower() else "audio"
-        in_rate = audio_fmt["input"]["format"]["rate"]
-        out_rate = (audio_fmt.get("output") or {}).get("format", {}).get("rate", 24000)
-
-        session = {"type": args.type}
-        if args.type == "realtime":
-            instructions = args.instructions
-            if instructions is None:
-                instructions = TOOLS_INSTRUCTIONS if args.tools else MODE_INSTRUCTIONS[card].get(args.output_modalities)
-            if args.tools:
-                session["tools"] = DEMO_TOOLS
-            if instructions is not None:
-                session["instructions"] = instructions
-            if args.output_modalities:
-                session["output_modalities"] = args.output_modalities.split(",")
-        elif args.language:
-            session["audio"] = {"input": {"transcription": {"language": args.language}}}
-        await conn.session.update(session=session)  # ty: ignore[invalid-argument-type]
-        ev = await conn.recv()
-        if ev.type == "error":
-            print("error:", ev.error.message)
-            return
-
-        turns = [("audio", path) for path in args.wav] + [("text", text) for text in args.text]
-        for i, (kind, src) in enumerate(turns):
-            t_last = 0.0
-
-            # the input is sent while the events are read: live transcription arrives while the audio is sent
-            async def send_turn():
-                nonlocal t_last
-                if kind == "audio":
-                    pcm = read_wav(src, in_rate)
-                    step = in_rate * args.chunk_ms // 1000
-                    for j in range(0, len(pcm), step):
-                        chunk = pcm[j:j + step]
-                        await conn.input_audio_buffer.append(audio=base64.b64encode(chunk.tobytes()).decode())
-                        if args.realtime:
-                            await asyncio.sleep(len(chunk) / in_rate)
-                    if args.type == "transcription":
-                        # "final" ends the input, models that transcribe with a delay then complete the last item
-                        await conn.send({"type": "input_audio_buffer.commit", "final": i == len(args.wav) - 1})  # ty: ignore[invalid-argument-type]
-                else:
-                    await conn.conversation.item.create(item={
-                        "type": "message", "role": "user", "content": [{"type": "input_text", "text": src}]})
-                if args.type == "realtime":
-                    # the audio not committed yet ends the user turn
-                    await conn.response.create()
-                t_last = time.perf_counter()
-
-            print(f"--- turn {i} ({kind}: {src})")
-            sender = asyncio.create_task(send_turn())
-            t_first = None
-            audio = []
-            async for ev in conn:
-                if ev.type == "response.output_text.delta" or ev.type == "conversation.item.input_audio_transcription.delta":
-                    t_first = t_first or time.perf_counter()
-                    print(ev.delta, end="", flush=True)
-                elif ev.type == "response.output_audio.delta":
-                    t_first = t_first or time.perf_counter()
-                    audio.append(np.frombuffer(base64.b64decode(ev.delta), dtype=np.int16))
-                elif ev.type == "response.done":
-                    # the calls of the response are run, their results go back and the model answers them
-                    calls = [o for o in ev.model_dump()["response"]["output"] if o["type"] == "function_call"]
-                    for call in calls:
-                        result = json.dumps(run_tool(call["name"], call["arguments"]))
-                        print(f"\n[call {call['name']}({call['arguments']}) -> {result}]", flush=True)
-                        await conn.conversation.item.create(item={
-                            "type": "function_call_output", "call_id": call["call_id"], "output": result})
-                    if not calls:
-                        break
-                    await conn.response.create()
-                elif ev.type == "conversation.item.input_audio_transcription.completed":
-                    break
-                elif ev.type == "error":
-                    print("\nerror:", ev.error.message)
-                    return
-            await sender
-            print()
-            if t_first is not None and t_first > t_last:
-                print(f"first output {1000 * (t_first - t_last):.0f} ms after the end of the input")
-            elif t_first is not None:
-                print(f"output started {1000 * (t_last - t_first):.0f} ms before the end of the input, "
-                      f"done {1000 * (time.perf_counter() - t_last):.0f} ms after it")
-            if audio:
-                pcm_out = np.concatenate(audio)
-                path = f"{args.out.removesuffix('.wav')}_{i}.wav" if len(turns) > 1 else args.out
-                write_wav(path, pcm_out, out_rate)
-                print(f"audio: {len(pcm_out) / out_rate:.2f} s -> {path}")
-
-
-if __name__ == "__main__":
-    p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
-    p.add_argument("wav", nargs="*", help="PCM16 wav files, one user turn each")
-    p.add_argument("--text", action="append", default=[], help="a user turn as text, after the audio turns, repeatable")
-    p.add_argument("--type", choices=["realtime", "transcription"], default="realtime", help="session type")
-    p.add_argument("--instructions", help="system prompt of the conversation")
-    p.add_argument("--tools", action="store_true", help="give the conversation a demo tool (set_volume), its calls are run")
-    p.add_argument("--output-modalities", choices=["text", "audio", "text,audio"], help="default: the model picks")
-    p.add_argument("--language", help="transcription language, e.g. en")
-    p.add_argument("--chunk-ms", type=int, default=100, help="audio chunk size")
-    p.add_argument("--realtime", action=argparse.BooleanOptionalAction, default=True, help="send the audio in real time")
-    p.add_argument("--out", default="response.wav", help="where the audio of the responses goes")
-    p.add_argument("--url", default="http://127.0.0.1:9931/v1")
-    p.add_argument("--api-key", default="none")
-    p.add_argument("--model", default="default", help="model name, used by the router")
-    asyncio.run(main(p.parse_args()))
@@ -4,7 +4,7 @@ project("ggml" C CXX ASM)

 ### GGML Version
 set(GGML_VERSION_MAJOR 0)
-set(GGML_VERSION_MINOR 28)
+set(GGML_VERSION_MINOR 27)
 set(GGML_VERSION_PATCH 0)
 set(GGML_VERSION_BASE "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")

@@ -250,7 +250,6 @@ option(GGML_OPENMP_FETCH                    "ggml: fetch LLVM OpenMP"
 option(GGML_RPC                             "ggml: use RPC"                                   OFF)
 option(GGML_SYCL                            "ggml: use SYCL"                                  OFF)
 option(GGML_SYCL_F16                        "ggml: use 16 bit floats for sycl calculations"   OFF)
-option(GGML_SYCL_SUPPORT_XMX                "ggml: build the XMX (DPAS) code paths in the SYCL backend" ON)
 option(GGML_SYCL_GRAPH                      "ggml: enable graphs in the SYCL backend"         ON)
 option(GGML_SYCL_HOST_MEM_FALLBACK          "ggml: allow host memory fallback in SYCL reorder (requires kernel 6.8+)" ON)
 option(GGML_SYCL_SUPPORT_LEVEL_ZERO_API     "ggml: use Level Zero API in SYCL backend"  ON)
@@ -70,8 +70,7 @@ extern "C" {
    GGML_API void                           ggml_backend_buffer_reset         (ggml_backend_buffer_t buffer);

    // tensor copy between different backends
-    GGML_API void ggml_backend_tensor_copy  (const struct ggml_tensor * src, struct ggml_tensor * dst);
-    GGML_API void ggml_backend_tensor_copy_n(const struct ggml_tensor * const * srcs, struct ggml_tensor * const * dsts, size_t n_tensors);
+    GGML_API void ggml_backend_tensor_copy(const struct ggml_tensor * src, struct ggml_tensor * dst);

    //
    // Backend (stream)
@@ -96,8 +95,6 @@ extern "C" {
    GGML_API void ggml_backend_tensor_get   (const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
    GGML_API void ggml_backend_tensor_set_2d(      struct ggml_tensor * tensor, const void * data, size_t offset, size_t size, size_t n_copies, size_t stride_tensor, size_t stride_data);
    GGML_API void ggml_backend_tensor_get_2d(const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size, size_t n_copies, size_t stride_tensor, size_t stride_data);
-    GGML_API void ggml_backend_tensor_set_n (      struct ggml_tensor * const * tensors, const void * data, const size_t * offsets, const size_t * sizes, const size_t * data_offsets, size_t n_tensors);
-    GGML_API void ggml_backend_tensor_get_n (const struct ggml_tensor * const * tensors,       void * data, const size_t * offsets, const size_t * sizes, const size_t * data_offsets, size_t n_tensors);
    GGML_API void ggml_backend_tensor_memset(      struct ggml_tensor * tensor,     uint8_t value, size_t offset, size_t size);

    GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
@@ -8,10 +8,10 @@ extern "C" {

 #define RPC_PROTO_MAJOR_VERSION    8
 #define RPC_PROTO_MINOR_VERSION    0
-#define RPC_PROTO_PATCH_VERSION    1
+#define RPC_PROTO_PATCH_VERSION    0

 #ifdef  __cplusplus
-static_assert(GGML_OP_COUNT == 102, "GGML_OP_COUNT has changed - update RPC_PROTO_PATCH_VERSION");
+static_assert(GGML_OP_COUNT == 101, "GGML_OP_COUNT has changed - update RPC_PROTO_PATCH_VERSION");
 #endif

 #define GGML_RPC_MAX_SERVERS       16
@@ -381,8 +381,6 @@ extern "C" {
    GGML_API void        ggml_fp32_to_bf16_row_ref(const float *, ggml_bf16_t *, int64_t);
    GGML_API void        ggml_fp32_to_bf16_row(const float *, ggml_bf16_t *, int64_t);

-    typedef struct { uint8_t bits; } ggml_fp8_e4m3_t;
-
    struct ggml_object;
    struct ggml_context;
    struct ggml_cgraph;
@@ -432,8 +430,7 @@ extern "C" {
        GGML_TYPE_NVFP4   = 40, // NVFP4 (4 blocks, E4M3 scale)
        GGML_TYPE_Q1_0    = 41,
        GGML_TYPE_Q2_0    = 42,
-        GGML_TYPE_F8_E4M3 = 43,
-        GGML_TYPE_COUNT   = 44,
+        GGML_TYPE_COUNT   = 43,
    };

    // [TAG_GGML_PREC]
@@ -489,7 +486,6 @@ extern "C" {
        GGML_FTYPE_MOSTLY_NVFP4   = 26, // except 1d tensors
        GGML_FTYPE_MOSTLY_Q1_0    = 27, // except 1d tensors
        GGML_FTYPE_MOSTLY_Q2_0    = 28, // except 1d tensors
-        GGML_FTYPE_MOSTLY_F8_E4M3 = 29, // except 1d tensors
    };

    // available tensor operations:
@@ -589,7 +585,6 @@ extern "C" {
        GGML_OP_DSV4_HC_COMB,
        GGML_OP_DSV4_HC_PRE,
        GGML_OP_DSV4_HC_POST,
-        GGML_OP_TOP_K_MASK,

        GGML_OP_UNARY,

@@ -930,7 +925,6 @@ extern "C" {
            enum   ggml_type      type);

    // dst[i0, i1, i2] = a[i0, i1, i2] + b[i0, ids[i1, i2]]
-    // an id of -1 adds nothing: dst[i0, i1, i2] = a[i0, i1, i2]
    GGML_API struct ggml_tensor * ggml_add_id(
            struct ggml_context * ctx,
            struct ggml_tensor  * a,
@@ -1482,16 +1476,6 @@ extern "C" {
            enum ggml_prec       prec,
            int                  idx);

-    // [TAG_GGML_PREC]
-    // get the accumulator / src[idx] precision set by the functions above
-    // return GGML_PREC_UNDEFINED when the op does not carry the requested precision
-    GGML_API enum ggml_prec ggml_prec_get_acc(
-            const struct ggml_tensor * a);
-
-    GGML_API enum ggml_prec ggml_prec_get_src(
-            const struct ggml_tensor * a,
-            int                        idx);
-
    // A: k columns, n rows => [ne03, ne02, n, k]
    // B: k columns, m rows  (i.e. we transpose it internally) => [ne03 * x, ne02 * y, m, k]
    // result is n columns, m rows => [ne03 * x, ne02 * y, m, n]
@@ -1513,7 +1497,6 @@ extern "C" {
            enum ggml_op_hint    hint);

    // indirect matrix multiplication
-    // an id of -1 skips the slot and zeroes the matching row of the result
    //
    // [TAG_MUL_MAT_ID_DUPLICATE_ID]
    // note: this op has undefined behaviour if a given `a` is referenced more than once for a given `b`
@@ -2749,26 +2732,6 @@ extern "C" {
            struct ggml_tensor  * post,
            struct ggml_tensor  * comb);

-    // top-k sparse attention mask: keep the mask cells of the live picks and the extra cells, set the rest to -inf
-    //
-    // mask:   [n_kv,    m1,  m2,  m3]  f16 or f32, broadcast to the rows of top_k
-    // top_k:  [n_top,   ne1, ne2, ne3] i32, picks of each row
-    // score:  [n_pick,  ne1, ne2, ne3] f32, can be NULL: a pick is live when its score is above -inf, else every pick is live
-    // blocks: [n_blk,   n_pick]        i32, can be NULL: the member cells of each pick, else a pick is a cell
-    // extra:  [n_extra, ne1, ne2, ne3] i32, can be NULL: cells of each row kept without a pick
-    // res:    [n_kv,    ne1, ne2, ne3] same type as mask
-    //
-    // res[c, i] = mask[c, i % m] when c is a cell of a live pick of row i or an extra cell of row i, else -inf
-    // picks and cells out of range are skipped
-    //
-    GGML_API struct ggml_tensor * ggml_top_k_mask(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * mask,
-            struct ggml_tensor  * top_k,
-            struct ggml_tensor  * score,
-            struct ggml_tensor  * blocks,
-            struct ggml_tensor  * extra);
-
    // custom operators

    typedef void (*ggml_custom1_op_t)(struct ggml_tensor * dst , const struct ggml_tensor * a, int ith, int nth, void * userdata);
@@ -47,9 +47,6 @@ extern "C" {
    // Backend buffer
    //

-    // TODO: remove `set_tensor`, `get_tensor`, `set_tensor_2d`, `get_tensor_2d` and `cpy_tensor` from the buffer interface
-    //       they are superseded by the `set_tensor_n`, `get_tensor_n`, `cpy_tensor_n` interface
-    //       ref: https://github.com/ggml-org/llama.cpp/pull/30881#discussion_r4400299490
    struct ggml_backend_buffer_i {
        // (optional) free the buffer
        void         (*free_buffer)  (ggml_backend_buffer_t buffer);
@@ -71,11 +68,6 @@ extern "C" {
        void         (*clear)        (ggml_backend_buffer_t buffer, uint8_t value);
        // (optional) reset any internal state due to tensor initialization, such as tensor extras
        void         (*reset)        (ggml_backend_buffer_t buffer);
-        // (optional) batched tensor data copies
-        void         (*set_tensor_n) (ggml_backend_buffer_t buffer,       struct ggml_tensor * const * tensors, const void * const * data, const size_t * offsets, const size_t * sizes, size_t n_tensors);
-        void         (*get_tensor_n) (ggml_backend_buffer_t buffer, const struct ggml_tensor * const * tensors,       void * const * data, const size_t * offsets, const size_t * sizes, size_t n_tensors);
-        // (optional) batched tensor copies: returns true if all copies were performed, false if unsupported and no copies were performed
-        bool         (*cpy_tensor_n) (ggml_backend_buffer_t buffer, const struct ggml_tensor * const * srcs, struct ggml_tensor * const * dsts, size_t n_tensors);
    };

    struct ggml_backend_buffer {
@@ -774,6 +774,13 @@ static struct ggml_backend_meta_split_state ggml_backend_meta_get_split_state(
        return src_ss[0];
    };

+    auto handle_pool_1d = [&](const std::vector<ggml_backend_meta_split_state> & src_ss) -> ggml_backend_meta_split_state {
+        if (src_ss[0].axis == GGML_BACKEND_SPLIT_AXIS_0) {
+            return {GGML_BACKEND_SPLIT_AXIS_MIRRORED, {0}, {1}, 1};
+        }
+        return src_ss[0];
+    };
+
    auto handle_flash_attn_ext = [&](const std::vector<ggml_backend_meta_split_state> & src_ss) -> ggml_backend_meta_split_state {
        GGML_ASSERT(tensor->src[3] == nullptr || src_ss[3].axis == GGML_BACKEND_SPLIT_AXIS_MIRRORED);

@@ -990,12 +997,14 @@ static struct ggml_backend_meta_split_state ggml_backend_meta_get_split_state(
            case GGML_OP_CONV_3D:
            case GGML_OP_CONV_2D_DW:
            case GGML_OP_CONV_TRANSPOSE_2D:
-            case GGML_OP_POOL_1D:
            case GGML_OP_POOL_2D:
            case GGML_OP_POOL_2D_BACK:
            case GGML_OP_UPSCALE: {
                split_state = handle_generic(src_ss, /*scalar_only =*/ true);
            } break;
+            case GGML_OP_POOL_1D: {
+                split_state = handle_pool_1d(src_ss);
+            } break;
            case GGML_OP_PAD: {
                split_state = handle_pad(src_ss);
            } break;
@@ -1046,8 +1055,7 @@ static struct ggml_backend_meta_split_state ggml_backend_meta_get_split_state(
            } break;
            case GGML_OP_DSV4_HC_COMB:
            case GGML_OP_DSV4_HC_PRE:
-            case GGML_OP_DSV4_HC_POST:
-            case GGML_OP_TOP_K_MASK: {
+            case GGML_OP_DSV4_HC_POST: {
                split_state = handle_generic(src_ss, /*scalar_only =*/ true);
            } break;
            case GGML_OP_UNARY: {
@@ -1699,9 +1707,6 @@ static const ggml_backend_buffer_i ggml_backend_meta_buffer_iface = {
    /* .cpy_tensor      = */ nullptr,
    /* .clear           = */ ggml_backend_meta_buffer_clear,
    /* .reset           = */ ggml_backend_meta_buffer_reset,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 bool ggml_backend_buffer_is_meta(ggml_backend_buffer_t buf) {
@@ -402,62 +402,6 @@ size_t ggml_backend_get_max_size(ggml_backend_t backend) {
    return ggml_backend_buft_get_max_size(ggml_backend_get_default_buffer_type(backend));
 }

-template <typename cb_get_buffer>
-static void ggml_backend_tensor_buft_groups(size_t n_tensors, cb_get_buffer && get_buffer, std::vector<std::vector<size_t>> & groups) {
-    std::unordered_map<ggml_backend_buffer_type_t, size_t> buft_idx;
-    groups.clear();
-    for (size_t i = 0; i < n_tensors; i++) {
-        ggml_backend_buffer_t buf = get_buffer(i);
-        auto it = buft_idx.find(buf->buft);
-        if (it == buft_idx.end()) {
-            buft_idx[buf->buft] = groups.size();
-            groups.emplace_back();
-        }
-        groups[buft_idx[buf->buft]].push_back(i);
-    }
-}
-
-static bool ggml_backend_tensor_batch_is_2d(
-        const struct ggml_tensor * const * tensors,
-        const size_t * offsets,
-        const size_t * sizes,
-        const size_t * data_offsets,
-        size_t n_tensors,
-        size_t * base_offset,
-        size_t * size,
-        size_t * stride_tensor,
-        size_t * stride_data) {
-    if (n_tensors <= 1) {
-        return false;
-    }
-
-    const struct ggml_tensor * tensor = tensors[0];
-    const size_t size0 = sizes[0];
-
-    if (size0 == 0 || offsets[1] < offsets[0] || data_offsets[1] < data_offsets[0]) {
-        return false;
-    }
-
-    const size_t stride_tensor0 = offsets[1] - offsets[0];
-    const size_t stride_data0   = data_offsets[1] - data_offsets[0];
-
-    for (size_t i = 1; i < n_tensors; i++) {
-        if (tensors[i] != tensor || sizes[i] != size0) {
-            return false;
-        }
-        if (offsets[i] != offsets[0] + i*stride_tensor0 || data_offsets[i] != data_offsets[0] + i*stride_data0) {
-            return false;
-        }
-    }
-
-    *base_offset   = offsets[0];
-    *size          = size0;
-    *stride_tensor = stride_tensor0;
-    *stride_data   = stride_data0;
-
-    return true;
-}
-
 void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
    GGML_ASSERT(backend);
    GGML_ASSERT(tensor);
@@ -602,154 +546,6 @@ void ggml_backend_tensor_get_2d(const struct ggml_tensor * tensor, void * data,
    buf->iface.get_tensor_2d(buf, tensor, data, offset, size, n_copies, stride_tensor, stride_data);
 }

-void ggml_backend_tensor_set_n(struct ggml_tensor * const * tensors, const void * data,
-        const size_t * offsets, const size_t * sizes, const size_t * data_offsets, size_t n_tensors) {
-    if (n_tensors == 0) {
-        return;
-    }
-
-    std::vector<const void *> data_ptrs(n_tensors);
-    for (size_t i = 0; i < n_tensors; i++) {
-        data_ptrs[i] = (const char *) data + data_offsets[i];
-    }
-
-    ggml_backend_buffer_t buf = NULL;
-    bool same_buft = true;
-
-    for (size_t i = 0; i < n_tensors; i++) {
-        GGML_ASSERT(tensors[i] != NULL);
-        ggml_backend_buffer_t tensor_buf = tensors[i]->view_src ? tensors[i]->view_src->buffer : tensors[i]->buffer;
-        GGML_ASSERT(tensor_buf != NULL && "tensor buffer not set");
-        GGML_ASSERT(tensors[i]->data != NULL && "tensor not allocated");
-        GGML_ASSERT(offsets[i] + sizes[i] <= ggml_nbytes(tensors[i]) && "tensor write out of bounds");
-
-        if (buf == NULL) {
-            buf = tensor_buf;
-        } else if (buf->buft != tensor_buf->buft) {
-            same_buft = false;
-        }
-    }
-
-    if (same_buft && buf->iface.set_tensor_n != NULL) {
-        buf->iface.set_tensor_n(buf, tensors, data_ptrs.data(), offsets, sizes, n_tensors);
-        return;
-    }
-
-    if (!same_buft) {
-        std::vector<std::vector<size_t>> groups;
-        ggml_backend_tensor_buft_groups(n_tensors, [&](size_t i) {
-            return tensors[i]->view_src ? tensors[i]->view_src->buffer : tensors[i]->buffer;
-        }, groups);
-
-        for (size_t g = 0; g < groups.size(); g++) {
-            std::vector<ggml_tensor *> g_tensors;
-            std::vector<size_t> g_offsets;
-            std::vector<size_t> g_sizes;
-            std::vector<size_t> g_data_offsets;
-            g_tensors.reserve(groups[g].size());
-            g_offsets.reserve(groups[g].size());
-            g_sizes.reserve(groups[g].size());
-            g_data_offsets.reserve(groups[g].size());
-
-            for (size_t i : groups[g]) {
-                g_tensors.push_back(tensors[i]);
-                g_offsets.push_back(offsets[i]);
-                g_sizes.push_back(sizes[i]);
-                g_data_offsets.push_back(data_offsets[i]);
-            }
-
-            ggml_backend_tensor_set_n(g_tensors.data(), data, g_offsets.data(), g_sizes.data(), g_data_offsets.data(), g_tensors.size());
-        }
-        return;
-    }
-
-    size_t base_offset = 0, size = 0, stride_tensor = 0, stride_data = 0;
-    if (ggml_backend_tensor_batch_is_2d(tensors, offsets, sizes, data_offsets, n_tensors,
-                &base_offset, &size, &stride_tensor, &stride_data)) {
-        ggml_backend_tensor_set_2d(tensors[0], data_ptrs[0], base_offset, size,
-                n_tensors, stride_tensor, stride_data);
-        return;
-    }
-
-    for (size_t i = 0; i < n_tensors; i++) {
-        ggml_backend_tensor_set(tensors[i], data_ptrs[i], offsets[i], sizes[i]);
-    }
-}
-
-void ggml_backend_tensor_get_n(const struct ggml_tensor * const * tensors, void * data,
-        const size_t * offsets, const size_t * sizes, const size_t * data_offsets, size_t n_tensors) {
-    if (n_tensors == 0) {
-        return;
-    }
-
-    std::vector<void *> data_ptrs(n_tensors);
-    for (size_t i = 0; i < n_tensors; i++) {
-        data_ptrs[i] = (char *) data + data_offsets[i];
-    }
-
-    ggml_backend_buffer_t buf = NULL;
-    bool same_buft = true;
-
-    for (size_t i = 0; i < n_tensors; i++) {
-        GGML_ASSERT(tensors[i] != NULL);
-        ggml_backend_buffer_t tensor_buf = tensors[i]->view_src ? tensors[i]->view_src->buffer : tensors[i]->buffer;
-        GGML_ASSERT(tensor_buf != NULL && "tensor buffer not set");
-        GGML_ASSERT(tensors[i]->data != NULL && "tensor not allocated");
-        GGML_ASSERT(offsets[i] + sizes[i] <= ggml_nbytes(tensors[i]) && "tensor read out of bounds");
-
-        if (buf == NULL) {
-            buf = tensor_buf;
-        } else if (buf->buft != tensor_buf->buft) {
-            same_buft = false;
-        }
-    }
-
-    if (same_buft && buf->iface.get_tensor_n != NULL) {
-        buf->iface.get_tensor_n(buf, tensors, data_ptrs.data(), offsets, sizes, n_tensors);
-        return;
-    }
-
-    if (!same_buft) {
-        std::vector<std::vector<size_t>> groups;
-        ggml_backend_tensor_buft_groups(n_tensors, [&](size_t i) {
-            return tensors[i]->view_src ? tensors[i]->view_src->buffer : tensors[i]->buffer;
-        }, groups);
-
-        for (size_t g = 0; g < groups.size(); g++) {
-            std::vector<const ggml_tensor *> g_tensors;
-            std::vector<size_t> g_offsets;
-            std::vector<size_t> g_sizes;
-            std::vector<size_t> g_data_offsets;
-            g_tensors.reserve(groups[g].size());
-            g_offsets.reserve(groups[g].size());
-            g_sizes.reserve(groups[g].size());
-            g_data_offsets.reserve(groups[g].size());
-
-            for (size_t i : groups[g]) {
-                g_tensors.push_back(tensors[i]);
-                g_offsets.push_back(offsets[i]);
-                g_sizes.push_back(sizes[i]);
-                g_data_offsets.push_back(data_offsets[i]);
-            }
-
-            ggml_backend_tensor_get_n(g_tensors.data(), data, g_offsets.data(), g_sizes.data(), g_data_offsets.data(), g_tensors.size());
-        }
-        return;
-    }
-
-    size_t base_offset = 0, size = 0, stride_tensor = 0, stride_data = 0;
-    if (ggml_backend_tensor_batch_is_2d(tensors, offsets, sizes, data_offsets, n_tensors,
-                &base_offset, &size, &stride_tensor, &stride_data)) {
-        ggml_backend_tensor_get_2d(tensors[0], data_ptrs[0], base_offset, size,
-                n_tensors, stride_tensor, stride_data);
-        return;
-    }
-
-    for (size_t i = 0; i < n_tensors; i++) {
-        ggml_backend_tensor_get(tensors[i], data_ptrs[i], offsets[i], sizes[i]);
-    }
-}
-
 void ggml_backend_tensor_memset(struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
    GGML_ASSERT(tensor);
    ggml_backend_buffer_t buf = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
@@ -852,59 +648,6 @@ void ggml_backend_tensor_copy(const struct ggml_tensor * src, struct ggml_tensor
    }
 }

-void ggml_backend_tensor_copy_n(const struct ggml_tensor * const * srcs, struct ggml_tensor * const * dsts, size_t n_tensors) {
-    if (n_tensors == 0) {
-        return;
-    }
-
-    for (size_t i = 0; i < n_tensors; ++i) {
-        GGML_ASSERT(ggml_are_same_layout(srcs[i], dsts[i]) && "cannot copy tensors with different layouts");
-    }
-
-    ggml_backend_buffer_t buf = dsts[0]->view_src ? dsts[0]->view_src->buffer : dsts[0]->buffer;
-    bool same_buft = true;
-    for (size_t i = 0; i < n_tensors; ++i) {
-        ggml_backend_buffer_t dst_buf = dsts[i]->view_src ? dsts[i]->view_src->buffer : dsts[i]->buffer;
-        if (dst_buf->buft != buf->buft) {
-            same_buft = false;
-            break;
-        }
-    }
-
-    // if cpy_tensor_n returns false, it must not have performed any copies
-    if (same_buft && buf->iface.cpy_tensor_n != NULL) {
-        if (buf->iface.cpy_tensor_n(buf, srcs, dsts, n_tensors)) {
-            return;
-        }
-    }
-
-    if (!same_buft) {
-        std::vector<std::vector<size_t>> groups;
-        ggml_backend_tensor_buft_groups(n_tensors, [&](size_t i) {
-            return dsts[i]->view_src ? dsts[i]->view_src->buffer : dsts[i]->buffer;
-        }, groups);
-
-        for (size_t g = 0; g < groups.size(); g++) {
-            std::vector<const ggml_tensor *> g_srcs;
-            std::vector<ggml_tensor *> g_dsts;
-            g_srcs.reserve(groups[g].size());
-            g_dsts.reserve(groups[g].size());
-
-            for (size_t i : groups[g]) {
-                g_srcs.push_back(srcs[i]);
-                g_dsts.push_back(dsts[i]);
-            }
-
-            ggml_backend_tensor_copy_n(g_srcs.data(), g_dsts.data(), g_srcs.size());
-        }
-        return;
-    }
-
-    for (size_t i = 0; i < n_tensors; ++i) {
-        ggml_backend_tensor_copy(srcs[i], dsts[i]);
-    }
-}
-
 void ggml_backend_tensor_copy_async(ggml_backend_t backend_src, ggml_backend_t backend_dst, const struct ggml_tensor * src, struct ggml_tensor * dst) {
    GGML_ASSERT(ggml_are_same_layout(src, dst) && "cannot copy tensors with different layouts");

@@ -1110,9 +853,6 @@ static const struct ggml_backend_buffer_i ggml_backend_multi_buffer_i = {
    /* .cpy_tensor      = */ NULL,
    /* .clear           = */ ggml_backend_multi_buffer_clear,
    /* .reset           = */ NULL,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 ggml_backend_buffer_t ggml_backend_multi_buffer_alloc_buffer(ggml_backend_buffer_t * buffers, size_t n_buffers) {
@@ -2792,9 +2532,6 @@ static const struct ggml_backend_buffer_i ggml_backend_cpu_buffer_i = {
    /* .cpy_tensor      = */ ggml_backend_cpu_buffer_cpy_tensor,
    /* .clear           = */ ggml_backend_cpu_buffer_clear,
    /* .reset           = */ NULL,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 static const struct ggml_backend_buffer_i ggml_backend_cpu_buffer_from_ptr_i = {
@@ -2809,9 +2546,6 @@ static const struct ggml_backend_buffer_i ggml_backend_cpu_buffer_from_ptr_i = {
    /* .cpy_tensor      = */ ggml_backend_cpu_buffer_cpy_tensor,
    /* .clear           = */ ggml_backend_cpu_buffer_clear,
    /* .reset           = */ NULL,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 // CPU backend buffer type
@@ -920,7 +920,7 @@ static void * ggml_backend_cann_buffer_get_base(ggml_backend_buffer_t buffer) {
 * @param dst Pointer to the destination buffer where transformed data will be
 * stored.
 */
-static void ggml_backend_cann_transform_q4_0(const ggml_tensor * tensor, const void * src, void * dst) {
+static void ggml_backend_cann_transform_q4_0(ggml_tensor * tensor, const void * src, void * dst) {
    int64_t n_elems     = ggml_nelements(tensor);
    int64_t groups      = n_elems / QK4_0;
    size_t  quant_bytes = n_elems * sizeof(uint8_t) / 2;
@@ -1014,7 +1014,7 @@ static void ggml_backend_cann_transform_back_q4_0(const ggml_tensor * tensor, vo
 * @param dst Pointer to the destination buffer where transformed data will be
 * stored.
 */
-static void ggml_backend_cann_transform_q8_0(const ggml_tensor * tensor, const void * src, void * dst) {
+static void ggml_backend_cann_transform_q8_0(ggml_tensor * tensor, const void * src, void * dst) {
    int64_t n_elems     = ggml_nelements(tensor);
    int64_t groups      = n_elems / QK8_0;
    size_t  quant_bytes = n_elems * sizeof(uint8_t);
@@ -1075,7 +1075,7 @@ static void ggml_backend_cann_transform_back_q8_0(const ggml_tensor * tensor, co
 * @param dst Pointer to the destination buffer where transformed data will be
 * stored.
 */
-static void ggml_backend_cann_transform(const ggml_tensor * tensor, const void * src, void * dst) {
+static void ggml_backend_cann_transform(ggml_tensor * tensor, const void * src, void * dst) {
    switch (tensor->type) {
        case GGML_TYPE_Q4_0:
            ggml_backend_cann_transform_q4_0(tensor, src, dst);
@@ -1478,9 +1478,6 @@ static const ggml_backend_buffer_i ggml_backend_cann_buffer_interface = {
    /* .cpy_tensor      = */ ggml_backend_cann_buffer_cpy_tensor,
    /* .clear           = */ ggml_backend_cann_buffer_clear,
    /* .reset           = */ NULL,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 /**
@@ -650,9 +650,9 @@ function(ggml_add_cpu_backend_variant_impl tag_name)

        # Fetch KleidiAI sources:
        include(FetchContent)
-        set(KLEIDIAI_COMMIT_TAG "v1.32.0")
+        set(KLEIDIAI_COMMIT_TAG "v1.31.0")
        set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/releases/download/${KLEIDIAI_COMMIT_TAG}/kleidiai-${KLEIDIAI_COMMIT_TAG}-src.tar.gz")
-        set(KLEIDIAI_RELEASE_ARCHIVE_MD5  "05f1953428ffe8e8c7f585dc6d508887")
+        set(KLEIDIAI_RELEASE_ARCHIVE_MD5  "930495f3aeb16b25dc18a2bde45924f4")

        set(KLEIDIAI_FETCH_ARGS
            URL ${KLEIDIAI_DOWNLOAD_URL}
@@ -116,9 +116,6 @@ static ggml_backend_buffer_i ggml_backend_amx_buffer_interface = {
    /* .cpy_tensor      = */ nullptr,
    /* .clear           = */ ggml_backend_amx_buffer_clear,
    /* .reset           = */ nullptr,
-    /* .set_tensor_n    = */ NULL,
-    /* .get_tensor_n    = */ NULL,
-    /* .cpy_tensor_n    = */ NULL,
 };

 static const char * ggml_backend_amx_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
@@ -287,9 +287,6 @@ void ggml_vec_dot_q2_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi

    sumf = vaddvq_f32(sumv);
 #else
-    UNUSED(nb);
-    UNUSED(x);
-    UNUSED(y);
    ggml_vec_dot_q2_0_q8_0_generic(n, s, bs, vx, bx, vy, by, nrc);
    return;
 #endif
@@ -535,60 +535,6 @@ void ggml_gemv_mxfp4_4x4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
    UNUSED(ncols_interleaved);
    UNUSED(blocklen);

-#if defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
-    if (svcntb() * 8 == 256) {
-        const svbool_t  pg16 = svptrue_pat_b8(SV_VL16);
-        const svbool_t  pg4  = svptrue_pat_b32(SV_VL4);
-        const svbool_t  pgb  = svptrue_b8();
-        const svbool_t  pgw  = svptrue_b32();
-        const svbool_t  pg_hi4 = svnot_b_z(pgw, pg4);
-        const svuint8_t m4b = svdup_n_u8(0x0f);
-        const svuint8_t values = svld1_u8(pg16, (const uint8_t *) kvalues_mxfp4);
-
-        // Reorder Q8 four-byte groups as [0,2,4,6,1,3,5,7]
-        const svuint32_t q8_group_index = svsub_n_u32_m(pg_hi4, svindex_u32(0, 2), 7);
-
-        const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
-        float * res_ptr = s;
-
-        for (int x = 0; x < nc / ncols_interleaved; x++) {
-            const block_mxfp4x4 * b_ptr = (const block_mxfp4x4 *) vx + (x * nb);
-            svfloat32_t sumf = svdup_n_f32(0.0f);
-
-            for (int l = 0; l < nb; l++) {
-                const svint8_t q8 = svld1_s8(pgb, a_ptr[l].qs);
-                const svint8_t q8_even_odd = svreinterpret_s8_u32(svtbl_u32(svreinterpret_u32_s8(q8), q8_group_index));
-                const svuint8_t qmx01 = svld1_u8(pgb, b_ptr[l].qs + 0);
-                const svuint8_t qmx_lo01 = svtbl_u8(values, svand_u8_x(pgb, qmx01, m4b));
-                const svuint8_t qmx_hi01 = svtbl_u8(values, svlsr_n_u8_x(pgb, qmx01, 4));
-                const svuint8_t qmx23 = svld1_u8(pgb, b_ptr[l].qs + 32);
-                const svuint8_t qmx_lo23 = svtbl_u8(values, svand_u8_x(pgb, qmx23, m4b));
-                const svuint8_t qmx_hi23 = svtbl_u8(values, svlsr_n_u8_x(pgb, qmx23, 4));
-
-                svint32_t sumi = svdup_n_s32(0);
-                sumi = svdot_lane_s32(sumi, svreinterpret_s8_u8(qmx_lo01), q8_even_odd, 0);
-                sumi = svdot_lane_s32(sumi, svreinterpret_s8_u8(qmx_lo23), q8_even_odd, 1);
-                sumi = svdot_lane_s32(sumi, svreinterpret_s8_u8(qmx_hi01), q8_even_odd, 2);
-                sumi = svdot_lane_s32(sumi, svreinterpret_s8_u8(qmx_hi23), q8_even_odd, 3);
-                const svint32_t sumi_cols = svadd_s32_z(pg4, sumi, svext_s32(sumi, sumi, 4));
-
-                // Vector form of GGML_CPU_E8M0_TO_FP32_HALF
-                const svuint32_t e_u32 = svld1ub_u32(pg4, b_ptr[l].e);
-                const svbool_t e_is_denormal = svcmplt_n_u32(pg4, e_u32, 2);
-                const svuint32_t e_denormal_bits = svlsl_u32_x(pg4, svdup_n_u32(0x00200000), e_u32);
-                const svuint32_t e_normal_bits = svlsl_n_u32_x(pg4, svsub_n_u32_x(pg4, e_u32, 1), 23);
-                const svuint32_t e_bits = svsel_u32(e_is_denormal, e_denormal_bits, e_normal_bits);
-                const svfloat16_t d_f16 = svreinterpret_f16_u16(svdup_n_u16(a_ptr[l].d));
-                const svfloat32_t d_f32 = svcvt_f32_f16_x(pg4, d_f16);
-                const svfloat32_t scale = svmul_f32_x(pg4, svreinterpret_f32_u32(e_bits), d_f32);
-                sumf = svmla_f32_m(pg4, sumf,svcvt_f32_s32_m(svdup_n_f32(0.0f), pg4, sumi_cols), scale);
-            }
-            svst1_f32(pg4, res_ptr + x * 4, sumf);
-        }
-        return;
-    }
-#endif // defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
-
 #if ! ((defined(_MSC_VER)) && ! defined(__clang__)) && defined(__aarch64__) && defined(__ARM_NEON) && defined(__ARM_FEATURE_DOTPROD)
    const int8x16_t kvalues = vld1q_s8(kvalues_mxfp4);
    const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
@@ -1845,71 +1791,6 @@ void ggml_gemv_q8_0_4x8_q8_0(int                        n,
    UNUSED(ncols_interleaved);
    UNUSED(blocklen);

-#if defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
-    // 256-bit SVE: 32 x int8 -> 8 x int32
-    if (ggml_cpu_get_sve_cnt() == QK8_0) {
-        const block_q8_0x4 * GGML_RESTRICT b_ptr = (const block_q8_0x4 *) vx;
-
-        const svbool_t pg   = svptrue_b8();
-        const svbool_t pg4w = svptrue_pat_b32(SV_VL4);
-
-        // after svtbl + svcvt the scale lanes are [d0,d0,d1,d1,d2,d2,d3,d3]: lane pair j = row j
-        const svuint16_t d_idx = svreinterpret_u16_u32(svlsr_n_u32_x(pg, svindex_u32(0, 1), 1));
-
-        for (int c = 0; c < nc; c += ncols_interleaved) {
-            const block_q8_0 * GGML_RESTRICT a_ptr = (const block_q8_0 *) vy;
-
-            svfloat32_t acc = svdup_n_f32(0.0f);
-
-            for (int b = 0; b < nb; ++b) {
-                const int8_t * GGML_RESTRICT bq = (const int8_t *) b_ptr->qs;
-
-                uint64_t a_chunk0;
-                uint64_t a_chunk1;
-                uint64_t a_chunk2;
-                uint64_t a_chunk3;
-                uint64_t b_d;
-
-                memcpy(&a_chunk0, a_ptr->qs +  0, sizeof(a_chunk0));
-                memcpy(&a_chunk1, a_ptr->qs +  8, sizeof(a_chunk1));
-                memcpy(&a_chunk2, a_ptr->qs + 16, sizeof(a_chunk2));
-                memcpy(&a_chunk3, a_ptr->qs + 24, sizeof(a_chunk3));
-                memcpy(&b_d, b_ptr->d, sizeof(b_d));
-
-                const svint8_t aq0 = svreinterpret_s8_u64(svdup_n_u64(a_chunk0));
-                const svint8_t aq1 = svreinterpret_s8_u64(svdup_n_u64(a_chunk1));
-                const svint8_t aq2 = svreinterpret_s8_u64(svdup_n_u64(a_chunk2));
-                const svint8_t aq3 = svreinterpret_s8_u64(svdup_n_u64(a_chunk3));
-
-                const svint8_t bq0 = svld1_s8(pg, bq +  0);
-                const svint8_t bq1 = svld1_s8(pg, bq + 32);
-                const svint8_t bq2 = svld1_s8(pg, bq + 64);
-                const svint8_t bq3 = svld1_s8(pg, bq + 96);
-
-                const svint32_t iacc0 = svdot_s32(svdup_n_s32(0), bq0, aq0);
-                const svint32_t iacc1 = svdot_s32(svdup_n_s32(0), bq1, aq1);
-                const svint32_t iacc2 = svdot_s32(svdup_n_s32(0), bq2, aq2);
-                const svint32_t iacc3 = svdot_s32(svdup_n_s32(0), bq3, aq3);
-
-                const svint32_t isum = svadd_s32_x(pg, svadd_s32_x(pg, iacc0, iacc1), svadd_s32_x(pg, iacc2, iacc3));
-
-                const svfloat16_t bd_h  = svtbl_f16(svreinterpret_f16_u64(svdup_n_u64(b_d)), d_idx);
-                const svfloat32_t scale = svmul_n_f32_x(pg, svcvt_f32_f16_x(pg, bd_h), GGML_CPU_FP16_TO_FP32(a_ptr->d));
-
-                acc = svmla_f32_x(pg, acc, svcvt_f32_s32_x(pg, isum), scale);
-
-                ++a_ptr;
-                ++b_ptr;
-            }
-
-            svst1_f32(pg4w, s, svadd_f32_x(pg, svuzp1_f32(acc, acc), svuzp2_f32(acc, acc)));
-            s += ncols_interleaved;
-        }
-
-        return;
-    }
-#endif  // defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
-
 #if defined(__aarch64__) && defined(__ARM_NEON) && defined(__ARM_FEATURE_DOTPROD)
    const block_q8_0x4 * b_ptr = (const block_q8_0x4 *) vx;

@@ -4993,7 +4874,6 @@ void ggml_gemm_q6_K_8x8_q8_K(int                        n,

    constexpr int ncols_interleaved = 8;
    constexpr int blocklen          = 8;
-    constexpr int q8_k_blocklen     = 4;

    assert(n % qk == 0);
    assert(nr % 4 == 0);
@@ -5002,195 +4882,9 @@ void ggml_gemm_q6_K_8x8_q8_K(int                        n,
    UNUSED(nb);
    UNUSED(ncols_interleaved);
    UNUSED(blocklen);
-    UNUSED(q8_k_blocklen);
-
-#if defined(__aarch64__) && defined(__ARM_FEATURE_SVE) && defined(__ARM_FEATURE_MATMUL_INT8)
-    if (svcntb() * 8 == 256) {
-        const svbool_t pg8  = svptrue_b8();
-        const svbool_t pg32 = svptrue_b32();
-
-        const svuint8_t low_nibble = svdup_n_u8(0x0f);
-        const svuint8_t high_lo2   = svdup_n_u8(0x03);
-        const svuint8_t high_hi2   = svdup_n_u8(0x30);
-        const svint8_t  q6_bias    = svdup_n_s8(32);
-        const svint32_t zero_s32   = svdup_n_s32(0);
-
-        for (int y = 0; y < nr / q8_k_blocklen; ++y) {
-            const block_q8_Kx4 * GGML_RESTRICT q8_ptr = (const block_q8_Kx4 *) vy + y * nb;
-
-            for (int x = 0; x < nc / ncols_interleaved; ++x) {
-                const block_q6_Kx8 * GGML_RESTRICT q6_ptr = (const block_q6_Kx8 *) vx + x * nb;
-
-                // one accumlator per output row
-                svfloat32_t acc_f32_r0 = svdup_n_f32(0);
-                svfloat32_t acc_f32_r1 = svdup_n_f32(0);
-                svfloat32_t acc_f32_r2 = svdup_n_f32(0);
-                svfloat32_t acc_f32_r3 = svdup_n_f32(0);
-
-                for (int b = 0; b < nb; ++b) {
-                    svint32_t acc_c03_r01 = zero_s32;
-                    svint32_t acc_c03_r23 = zero_s32;
-                    svint32_t acc_c47_r01 = zero_s32;
-                    svint32_t acc_c47_r23 = zero_s32;
-
-                    for (int half = 0; half < 2; ++half) {
-                        const uint8_t * ql_base = q6_ptr[b].ql + half * 512;
-                        const uint8_t * qh_base = q6_ptr[b].qh + half * 256;
-
-                        for (int sb = 0; sb < QK_K / 64; ++sb) {
-                            const int8_t * q8_base_l = q8_ptr[b].qs + half * 512 + sb * 64;
-                            const int8_t * q8_base_h = q8_ptr[b].qs + half * 512 + 256 + sb * 64;
-
-                            // load 16 bytes (8B of R0, 8B of R1) and repeat in upper half
-                            // low lanes will multiply with 16 bytes of C0 and high lanes will multiply with 16 bytes of C1
-                            const svint8_t q8_l_r01_k0 = svld1rq_s8(pg8, q8_base_l + 0);
-                            const svint8_t q8_l_r01_k1 = svld1rq_s8(pg8, q8_base_l + 32);
-                            const svint8_t q8_l_r23_k0 = svld1rq_s8(pg8, q8_base_l + 16);
-                            const svint8_t q8_l_r23_k1 = svld1rq_s8(pg8, q8_base_l + 48);
-
-                            const svint8_t q8_h_r01_k0 = svld1rq_s8(pg8, q8_base_h + 0);
-                            const svint8_t q8_h_r01_k1 = svld1rq_s8(pg8, q8_base_h + 32);
-                            const svint8_t q8_h_r23_k0 = svld1rq_s8(pg8, q8_base_h + 16);
-                            const svint8_t q8_h_r23_k1 = svld1rq_s8(pg8, q8_base_h + 48);
-
-                            const int ql_off = sb * QK_K / 2;
-                            const int qh_off = ql_off & 255;
-
-                            const int scale_idx_l = half * 8 + sb;
-                            const int scale_idx_h = scale_idx_l + 4;
-
-                            const svint32_t scales_l_all = svld1sb_s32(pg32, q6_ptr[b].scales + scale_idx_l * 8);
-                            const svint32_t scales_h_all = svld1sb_s32(pg32, q6_ptr[b].scales + scale_idx_h * 8);
-
-                            const svint32_t scales_l_c03 = svzip1_s32(scales_l_all, scales_l_all);
-                            const svint32_t scales_h_c03 = svzip1_s32(scales_h_all, scales_h_all);
-
-                            {
-                                svuint8_t q6_ql0_01 = svld1_u8(pg8, ql_base + ql_off);
-                                svuint8_t q6_ql1_01 = svld1_u8(pg8, ql_base + ql_off + 64);
-                                svuint8_t q6_qh0_01 = svld1_u8(pg8, qh_base + qh_off);
-                                svuint8_t q6_qh1_01 = svld1_u8(pg8, qh_base + qh_off + 64);
-
-                                if (sb > 1) {
-                                    q6_qh0_01 = svlsr_n_u8_x(pg8, q6_qh0_01, 2);
-                                    q6_qh1_01 = svlsr_n_u8_x(pg8, q6_qh1_01, 2);
-                                }
-
-                                const svuint8_t q6_qh0_01_hi = svand_u8_x(pg8, q6_qh0_01, high_hi2);
-                                const svuint8_t q6_qh1_01_hi = svand_u8_x(pg8, q6_qh1_01, high_hi2);
-
-                                const svuint8_t q6_l0_01_bits = svorr_u8_x(pg8, svand_u8_x(pg8, q6_ql0_01, low_nibble), svlsl_n_u8_x(pg8, svand_u8_x(pg8, q6_qh0_01, high_lo2), 4));
-                                const svuint8_t q6_l1_01_bits = svorr_u8_x(pg8, svand_u8_x(pg8, q6_ql1_01, low_nibble), svlsl_n_u8_x(pg8, svand_u8_x(pg8, q6_qh1_01, high_lo2), 4));
-                                const svuint8_t q6_h0_01_bits = svorr_u8_x(pg8, svlsr_n_u8_x(pg8, q6_ql0_01, 4), q6_qh0_01_hi);
-                                const svuint8_t q6_h1_01_bits = svorr_u8_x(pg8, svlsr_n_u8_x(pg8, q6_ql1_01, 4), q6_qh1_01_hi);
-
-                                const svint8_t q6_l0_01 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_l0_01_bits), q6_bias);
-                                const svint8_t q6_l1_01 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_l1_01_bits), q6_bias);
-                                const svint8_t q6_h0_01 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_h0_01_bits), q6_bias);
-                                const svint8_t q6_h1_01 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_h1_01_bits), q6_bias);
-
-                                svint32_t dot_l_r01 = svmmla_s32(zero_s32, q6_l0_01, q8_l_r01_k0);
-                                dot_l_r01 = svmmla_s32(dot_l_r01, q6_l1_01, q8_l_r01_k1);
-
-                                svint32_t dot_h_r01 = svmmla_s32(zero_s32, q6_h0_01, q8_h_r01_k0);
-                                dot_h_r01 = svmmla_s32(dot_h_r01, q6_h1_01, q8_h_r01_k1);
-
-                                svint32_t dot_l_r23 = svmmla_s32(zero_s32, q6_l0_01, q8_l_r23_k0);
-                                dot_l_r23 = svmmla_s32(dot_l_r23, q6_l1_01, q8_l_r23_k1);
-
-                                svint32_t dot_h_r23 = svmmla_s32(zero_s32, q6_h0_01, q8_h_r23_k0);
-                                dot_h_r23 = svmmla_s32(dot_h_r23, q6_h1_01, q8_h_r23_k1);
-
-                                acc_c03_r01 = svmla_s32_x(pg32, acc_c03_r01, dot_l_r01, scales_l_c03);
-                                acc_c03_r01 = svmla_s32_x(pg32, acc_c03_r01, dot_h_r01, scales_h_c03);
-                                acc_c03_r23 = svmla_s32_x(pg32, acc_c03_r23, dot_l_r23, scales_l_c03);
-                                acc_c03_r23 = svmla_s32_x(pg32, acc_c03_r23, dot_h_r23, scales_h_c03);
-                            }
-
-                            const svint32_t scales_l_c47 = svzip2_s32(scales_l_all, scales_l_all);
-                            const svint32_t scales_h_c47 = svzip2_s32(scales_h_all, scales_h_all);
-
-                            {
-                                svuint8_t q6_ql0_23 = svld1_u8(pg8, ql_base + ql_off + 32);
-                                svuint8_t q6_ql1_23 = svld1_u8(pg8, ql_base + ql_off + 96);
-                                svuint8_t q6_qh0_23 = svld1_u8(pg8, qh_base + qh_off + 32);
-                                svuint8_t q6_qh1_23 = svld1_u8(pg8, qh_base + qh_off + 96);
-
-                                if (sb > 1) {
-                                    q6_qh0_23 = svlsr_n_u8_x(pg8, q6_qh0_23, 2);
-                                    q6_qh1_23 = svlsr_n_u8_x(pg8, q6_qh1_23, 2);
-                                }
-
-                                const svuint8_t q6_qh0_23_hi = svand_u8_x(pg8, q6_qh0_23, high_hi2);
-                                const svuint8_t q6_qh1_23_hi = svand_u8_x(pg8, q6_qh1_23, high_hi2);
-
-                                const svuint8_t q6_l0_23_bits = svorr_u8_x(pg8, svand_u8_x(pg8, q6_ql0_23, low_nibble), svlsl_n_u8_x(pg8, svand_u8_x(pg8, q6_qh0_23, high_lo2), 4));
-                                const svuint8_t q6_l1_23_bits = svorr_u8_x(pg8, svand_u8_x(pg8, q6_ql1_23, low_nibble), svlsl_n_u8_x(pg8, svand_u8_x(pg8, q6_qh1_23, high_lo2), 4));
-                                const svuint8_t q6_h0_23_bits = svorr_u8_x(pg8, svlsr_n_u8_x(pg8, q6_ql0_23, 4), q6_qh0_23_hi);
-                                const svuint8_t q6_h1_23_bits = svorr_u8_x(pg8, svlsr_n_u8_x(pg8, q6_ql1_23, 4), q6_qh1_23_hi);
-
-                                const svint8_t q6_l0_23 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_l0_23_bits), q6_bias);
-                                const svint8_t q6_l1_23 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_l1_23_bits), q6_bias);
-                                const svint8_t q6_h0_23 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_h0_23_bits), q6_bias);
-                                const svint8_t q6_h1_23 = svsub_s8_x(pg8, svreinterpret_s8_u8(q6_h1_23_bits), q6_bias);
-
-                                svint32_t dot_l_r01 = svmmla_s32(zero_s32, q6_l0_23, q8_l_r01_k0);
-                                dot_l_r01 = svmmla_s32(dot_l_r01, q6_l1_23, q8_l_r01_k1);
-
-                                svint32_t dot_h_r01 = svmmla_s32(zero_s32, q6_h0_23, q8_h_r01_k0);
-                                dot_h_r01 = svmmla_s32(dot_h_r01, q6_h1_23, q8_h_r01_k1);
-
-                                svint32_t dot_l_r23 = svmmla_s32(zero_s32, q6_l0_23, q8_l_r23_k0);
-                                dot_l_r23 = svmmla_s32(dot_l_r23, q6_l1_23, q8_l_r23_k1);
-
-                                svint32_t dot_h_r23 = svmmla_s32(zero_s32, q6_h0_23, q8_h_r23_k0);
-                                dot_h_r23 = svmmla_s32(dot_h_r23, q6_h1_23, q8_h_r23_k1);
-
-                                acc_c47_r01 = svmla_s32_x(pg32, acc_c47_r01, dot_l_r01, scales_l_c47);
-                                acc_c47_r01 = svmla_s32_x(pg32, acc_c47_r01, dot_h_r01, scales_h_c47);
-                                acc_c47_r23 = svmla_s32_x(pg32, acc_c47_r23, dot_l_r23, scales_l_c47);
-                                acc_c47_r23 = svmla_s32_x(pg32, acc_c47_r23, dot_h_r23, scales_h_c47);
-                            }
-                        }
-                    }
-
-                    // acc_c03_r01 -> c0-r0,c0-r1,c1-r0,c1-r1,c2-r0,c2-r1,c3-r0,c3-r1
-                    // acc_c47_r01 -> c4-r0,c4-r1,c5-r0,c5-r1,c6-r0,c6-r1,c7-r0,c7-r1
-                    // svuzp1_s32 picks even lanes -> c0-r0,c1-r0,c2-r0,c3-r0,c4-r0,c5-r0,c6-r0,c7-r0
-                    // similarly svuzp2_s32 picks remaining values from odd lanes
-                    const svint32_t dot_r0 = svuzp1_s32(acc_c03_r01, acc_c47_r01);
-                    const svint32_t dot_r1 = svuzp2_s32(acc_c03_r01, acc_c47_r01);
-                    const svint32_t dot_r2 = svuzp1_s32(acc_c03_r23, acc_c47_r23);
-                    const svint32_t dot_r3 = svuzp2_s32(acc_c03_r23, acc_c47_r23);
-
-                    const svuint32_t q6_d_bits = svld1uh_u32(pg32, (const uint16_t *) q6_ptr[b].d);
-                    const svfloat32_t q6_d_f32 = svcvt_f32_f16_x(pg32, svreinterpret_f16_u32(q6_d_bits));
-
-                    const svfloat32_t scale_r0 = svmul_n_f32_x(pg32, q6_d_f32, q8_ptr[b].d[0]);
-                    const svfloat32_t scale_r1 = svmul_n_f32_x(pg32, q6_d_f32, q8_ptr[b].d[1]);
-                    const svfloat32_t scale_r2 = svmul_n_f32_x(pg32, q6_d_f32, q8_ptr[b].d[2]);
-                    const svfloat32_t scale_r3 = svmul_n_f32_x(pg32, q6_d_f32, q8_ptr[b].d[3]);
-
-                    acc_f32_r0 = svmla_f32_x(pg32, acc_f32_r0, svcvt_f32_s32_x(pg32, dot_r0), scale_r0);
-                    acc_f32_r1 = svmla_f32_x(pg32, acc_f32_r1, svcvt_f32_s32_x(pg32, dot_r1), scale_r1);
-                    acc_f32_r2 = svmla_f32_x(pg32, acc_f32_r2, svcvt_f32_s32_x(pg32, dot_r2), scale_r2);
-                    acc_f32_r3 = svmla_f32_x(pg32, acc_f32_r3, svcvt_f32_s32_x(pg32, dot_r3), scale_r3);
-                }
-
-                const int row = y * q8_k_blocklen;
-                const int col = x * ncols_interleaved;
-
-                svst1_f32(pg32, s + (row + 0) * bs + col, acc_f32_r0);
-                svst1_f32(pg32, s + (row + 1) * bs + col, acc_f32_r1);
-                svst1_f32(pg32, s + (row + 2) * bs + col, acc_f32_r2);
-                svst1_f32(pg32, s + (row + 3) * bs + col, acc_f32_r3);
-            }
-        }
-        return;
-    }
-#endif  // SVE compile-time end

 #if defined(__aarch64__) && defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8)
+    constexpr int    q8_k_blocklen = 4;
    const uint8x16_t m4b           = vdupq_n_u8(0x0f);
    const uint8x16_t mask_lo       = vdupq_n_u8(0x03);
    const uint8x16_t mask_hi       = vdupq_n_u8(0x30);
@@ -43,14 +43,6 @@ static inline float f32_to_f32(float x) {
    return x;
 }

-static inline float f8_e4m3_to_f32(ggml_fp8_e4m3_t x) {
-    return ggml_f8_e4m3_to_fp32(x.bits);
-}
-
-static inline ggml_fp8_e4m3_t f32_to_f8_e4m3(float x) {
-    return { ggml_fp32_to_f8_e4m3(x) };
-}
-
 // TODO - merge this into the traits table, after using row-based conversions
 template <class T>
 struct type_conversion_table;
@@ -73,12 +65,6 @@ struct type_conversion_table<ggml_bf16_t> {
    static constexpr ggml_bf16_t (*from_f32)(float) = f32_to_bf16;
 };

-template <>
-struct type_conversion_table<ggml_fp8_e4m3_t> {
-    static constexpr float (*to_f32)(ggml_fp8_e4m3_t) = f8_e4m3_to_f32;
-    static constexpr ggml_fp8_e4m3_t (*from_f32)(float) = f32_to_f8_e4m3;
-};
-
 template <>
 struct type_conversion_table<int32_t> {
    static constexpr float (*to_f32)(int32_t) = i32_to_f32;
@@ -86,9 +86,6 @@ float ggml_table_f32_e8m0_half[1 << 8];
 // precomputed f32 table for ue4m3 (1 KB) (simd-mappings.h)
 float ggml_table_f32_ue4m3[1 << 8];

-// Signed FP8 needs its own table: UE4M3 ignores the sign, halves values for NVFP4, and maps NaN to zero.
-float ggml_table_f32_f8_e4m3[1 << 8];
-
 #if defined(__ARM_ARCH)
 struct ggml_arm_arch_features_type {
    int sve_cnt;
@@ -299,12 +296,6 @@ static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
        .vec_dot_type             = GGML_TYPE_Q8_0,
        .nrows                    = 1,
    },
-    [GGML_TYPE_F8_E4M3] = {
-        .from_float               = quantize_row_f8_e4m3,
-        .vec_dot                  = ggml_vec_dot_f8_e4m3_f32,
-        .vec_dot_type             = GGML_TYPE_F32,
-        .nrows                    = 1,
-    },
    [GGML_TYPE_Q2_K] = {
        .from_float               = quantize_row_q2_K,
        .vec_dot                  = ggml_vec_dot_q2_K_q8_K,
@@ -1661,12 +1652,7 @@ static void ggml_compute_forward_mul_mat_id(
            for (int id = 0; id < n_ids; ++id) {
                const int32_t i02 = *(const int32_t *) ((const char *) ids->data + iid1*ids->nb[1] + id*ids->nb[0]);

-                assert(i02 == -1 || (i02 >= 0 && i02 < n_as));
-
-                if (i02 == -1) {
-                    memset((char *) dst->data + id*nb1 + iid1*nb2, 0, ne0*sizeof(float));
-                    continue;
-                }
+                assert(i02 >= 0 && i02 < n_as);

                MMID_MATRIX_ROW(i02, matrix_row_counts[i02]) = (struct mmid_row_mapping) {id, iid1};
                matrix_row_counts[i02] += 1;
@@ -2119,10 +2105,6 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
            {
                ggml_compute_forward_dsv4_hc_post(params, tensor);
            } break;
-        case GGML_OP_TOP_K_MASK:
-            {
-                ggml_compute_forward_top_k_mask(params, tensor);
-            } break;
        case GGML_OP_MAP_CUSTOM1:
            {
                ggml_compute_forward_map_custom1(params, tensor);
@@ -2306,7 +2288,6 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
        case GGML_OP_DSV4_HC_COMB:
        case GGML_OP_DSV4_HC_PRE:
        case GGML_OP_DSV4_HC_POST:
-        case GGML_OP_TOP_K_MASK:
            {
                n_tasks = n_threads;
            } break;
@@ -2885,7 +2866,7 @@ struct ggml_cplan ggml_graph_plan(
                case GGML_OP_ADD_ID:
                case GGML_OP_ADD1:
                    {
-                        if (ggml_is_quantized(node->src[0]->type) || node->src[0]->type == GGML_TYPE_F8_E4M3) {
+                        if (ggml_is_quantized(node->src[0]->type)) {
                            cur = ggml_type_size(GGML_TYPE_F32) * node->src[0]->ne[0] * n_tasks;
                        }
                    } break;
@@ -2933,7 +2914,7 @@ struct ggml_cplan ggml_graph_plan(
                    } break;
                case GGML_OP_OUT_PROD:
                    {
-                        if (ggml_is_quantized(node->src[0]->type) || node->src[0]->type == GGML_TYPE_F8_E4M3 ||
+                        if (ggml_is_quantized(node->src[0]->type) ||
                            node->src[0]->type == GGML_TYPE_F16) {
                            cur = ggml_type_size(GGML_TYPE_F32) * node->src[0]->ne[0] * n_tasks;
                        }
@@ -3917,7 +3898,6 @@ void ggml_cpu_init(void) {
            // initialize UE4M3 table (256 entries)
            for (int i = 0; i < (1 << 8); ++i) {
                ggml_table_f32_ue4m3[i] = ggml_ue4m3_to_fp32(i);
-                ggml_table_f32_f8_e4m3[i] = ggml_f8_e4m3_to_fp32(i);
            }

            const uint64_t t_end = ggml_time_us(); UNUSED(t_end);
@@ -201,117 +201,85 @@ static struct kai_matmul_uker_config q4_k_matmul_config() {
    return config;
 }

-static struct kai_matmul_uker_config default_matmul_config() {
-    return {};
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static size_t q4_k_get_m_step() {
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    return api.get_step(&config).m;
 }

-template<
-        struct kai_matmul_uker_api (*GetApi)(void),
-        struct kai_matmul_uker_config (*GetConfig)(void),
-        size_t MR, size_t NR, size_t KR, size_t SR>
-struct matmul_uker_adapter {
-    static size_t get_m_step() {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        return api.get_step(&config).m;
-    }
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static size_t q4_k_get_n_step() {
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    return api.get_step(&config).n;
+}

-    static size_t get_n_step() {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        return api.get_step(&config).n;
-    }
+template<size_t Value>
+static size_t q4_k_get_constant() {
+    return Value;
+}

-    static size_t get_mr() {
-        if constexpr (MR == 0) {
-            return get_m_step();
-        }
-        return MR;
-    }
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static size_t q4_k_get_dst_offset(size_t m_idx, size_t n_idx, size_t dst_stride) {
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    const struct kai_matmul_uker_dst_dim_args index = { m_idx, n_idx };
+    const struct kai_matmul_uker_dst_stride_args stride = { dst_stride };
+    return api.get_dst_offset(&config, &index, &stride);
+}

-    static size_t get_nr() {
-        if constexpr (NR == 0) {
-            return get_n_step();
-        }
-        return NR;
-    }
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static size_t q4_k_get_lhs_packed_offset(size_t m_idx, size_t k, size_t bl) {
+    GGML_ASSERT(bl == KLEIDIAI_Q4_K_BLOCK_LENGTH);

-    static size_t get_kr() {
-        return KR;
-    }
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    const struct kai_matmul_uker_lhs_dim_args shape = { m_idx, k };
+    const struct kai_matmul_uker_lhs_dim_args index = { m_idx, 0 };
+    const struct kai_matmul_uker_lhs_stride_args stride = api.get_lhs_stride(&config, &shape);
+    return api.get_lhs_offset(&config, &index, &stride);
+}

-    static size_t get_sr() {
-        return SR;
-    }
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static size_t q4_k_get_rhs_packed_offset(size_t n_idx, size_t k, size_t bl) {
+    GGML_ASSERT(bl == KLEIDIAI_Q4_K_BLOCK_LENGTH);

-    static size_t get_dst_offset(size_t m_idx, size_t n_idx, size_t dst_stride) {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        const struct kai_matmul_uker_dst_dim_args index = { m_idx, n_idx };
-        const struct kai_matmul_uker_dst_stride_args stride = { dst_stride };
-        return api.get_dst_offset(&config, &index, &stride);
-    }
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    const struct kai_matmul_uker_rhs_dim_args shape = { n_idx, k };
+    const struct kai_matmul_uker_rhs_dim_args index = { n_idx, 0 };
+    const struct kai_matmul_uker_rhs_stride_args stride = api.get_rhs_stride(&config, &shape);
+    return api.get_rhs_offset(&config, &index, &stride);
+}

-    static size_t get_lhs_packed_offset(size_t m_idx, size_t k, size_t bl) {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        GGML_ASSERT(bl == config.format.bl);
-        const struct kai_matmul_uker_lhs_dim_args shape = { m_idx, k };
-        const struct kai_matmul_uker_lhs_dim_args index = { m_idx, 0 };
-        const struct kai_matmul_uker_lhs_stride_args stride = api.get_lhs_stride(&config, &shape);
-        return api.get_lhs_offset(&config, &index, &stride);
-    }
+template<struct kai_matmul_uker_api (*GetApi)(void)>
+static void q4_k_run_matmul(
+        size_t m, size_t n, size_t k, size_t bl,
+        const void * lhs_packed, const void * rhs_packed,
+        void * dst, size_t dst_stride_row, size_t dst_stride_col,
+        float clamp_min, float clamp_max) {
+    GGML_ASSERT(bl == KLEIDIAI_Q4_K_BLOCK_LENGTH);
+    GGML_ASSERT(dst_stride_col == sizeof(float));

-    static size_t get_rhs_packed_offset(size_t n_idx, size_t k, size_t bl) {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        GGML_ASSERT(bl == config.format.bl);
-        const struct kai_matmul_uker_rhs_dim_args shape = { n_idx, k };
-        const struct kai_matmul_uker_rhs_dim_args index = { n_idx, 0 };
-        const struct kai_matmul_uker_rhs_stride_args stride = api.get_rhs_stride(&config, &shape);
-        return api.get_rhs_offset(&config, &index, &stride);
-    }
+    const struct kai_matmul_uker_api api = GetApi();
+    const struct kai_matmul_uker_config config = q4_k_matmul_config();
+    const struct kai_matmul_uker_lhs_dim_args lhs_shape = { m, k };
+    const struct kai_matmul_uker_rhs_dim_args rhs_shape = { n, k };

-    static void run(
-            size_t m, size_t n, size_t k, size_t bl,
-            const void * lhs_packed, const void * rhs_packed,
-            void * dst, size_t dst_stride_row, size_t dst_stride_col,
-            float clamp_min, float clamp_max) {
-        const struct kai_matmul_uker_api api = GetApi();
-        const struct kai_matmul_uker_config config = GetConfig();
-        GGML_ASSERT(bl == config.format.bl);
-        GGML_ASSERT(dst_stride_col == sizeof(float));
-        const struct kai_matmul_uker_lhs_dim_args lhs_shape = { m, k };
-        const struct kai_matmul_uker_rhs_dim_args rhs_shape = { n, k };
-
-        struct kai_matmul_uker_args args = {};
-        args.flags = KAI_MATMUL_UKER_FLAGS_ARGS_CLAMP;
-        args.shape = { m, n, k };
-        args.operand.lhs.ptr = lhs_packed;
-        args.operand.lhs.stride = api.get_lhs_stride(&config, &lhs_shape);
-        args.operand.rhs.ptr = rhs_packed;
-        args.operand.rhs.stride = api.get_rhs_stride(&config, &rhs_shape);
-        args.operand.dst.ptr = dst;
-        args.operand.dst.stride.m = dst_stride_row;
-        args.activation.clamp.min_ptr = &clamp_min;
-        args.activation.clamp.max_ptr = &clamp_max;
-        api.run(&config, &args);
-    }
-};
-
-using q4_k_gemm_adapter = matmul_uker_adapter<
-    kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa,
-    q4_k_matmul_config, 0, 0, 4, 2>;
-using q4_k_gemv_adapter = matmul_uker_adapter<
-    kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot,
-    q4_k_matmul_config, 0, 0, 4, 2>;
-// GEMM and GEMV share the RHS layout only at SVE256 (NR=8, KR=8, SR=1).
-using sve_q8_gemm_adapter = matmul_uker_adapter<
-    kai_matmul_clamp_f32_qai8dxp4x8sf32_qsi8cxp4vsx8sf32bf32_16x4vs_sve_i8mm,
-    default_matmul_config, 4, 0, 8, 1>;
-using sve_q8_gemv_adapter = matmul_uker_adapter<
-    kai_matmul_clamp_f32_qai8dxp1x8_qsi8cxp8x8_1x8_sve_dot,
-    default_matmul_config, 1, 8, 8, 1>;
+    struct kai_matmul_uker_args args = {};
+    args.flags = KAI_MATMUL_UKER_FLAGS_ARGS_CLAMP;
+    args.shape = { m, n, k };
+    args.operand.lhs.ptr = lhs_packed;
+    args.operand.lhs.stride = api.get_lhs_stride(&config, &lhs_shape);
+    args.operand.rhs.ptr = rhs_packed;
+    args.operand.rhs.stride = api.get_rhs_stride(&config, &rhs_shape);
+    args.operand.dst.ptr = dst;
+    args.operand.dst.stride.m = dst_stride_row;
+    args.activation.clamp.min_ptr = &clamp_min;
+    args.activation.clamp.max_ptr = &clamp_max;
+    api.run(&config, &args);
+}

 static struct kai_matmul_pack_lhs_uker_config q4_k_lhs_pack_config() {
    struct kai_matmul_pack_lhs_uker_config config = {};
@@ -660,7 +628,7 @@ static ggml_kleidiai_kernels gemm_gemv_kernels[] = {
            /* .packed_size_ex        = */ &rhs_ps_fn5<kai_get_rhs_packed_size_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0>,
            /* .pack_func_ex          = */ &rhs_pack_fn12<kai_run_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0>,
        },
-        /* .required_cpu       = */ CPU_FEATURE_SVE | CPU_FEATURE_SVEI8MM | CPU_FEATURE_DOTPROD,
+        /* .required_cpu       = */ CPU_FEATURE_SVE | CPU_FEATURE_I8MM | CPU_FEATURE_DOTPROD,
        /* .lhs_type           = */ GGML_TYPE_F32,
        /* .rhs_type           = */ GGML_TYPE_Q4_0,
        /* .op_type            = */ GGML_TYPE_F32,
@@ -769,16 +737,16 @@ static ggml_kleidiai_kernels gemm_gemv_kernels_q4_k[] = {
    {
        /* SME2 GEMM */
        {
-            /* .get_m_step            = */ q4_k_gemm_adapter::get_m_step,
-            /* .get_n_step            = */ q4_k_gemm_adapter::get_n_step,
-            /* .get_mr                = */ q4_k_gemm_adapter::get_mr,
-            /* .get_nr                = */ q4_k_gemm_adapter::get_nr,
-            /* .get_kr                = */ q4_k_gemm_adapter::get_kr,
-            /* .get_sr                = */ q4_k_gemm_adapter::get_sr,
-            /* .get_dst_offset        = */ q4_k_gemm_adapter::get_dst_offset,
-            /* .get_lhs_offset_ex     = */ q4_k_gemm_adapter::get_lhs_packed_offset,
-            /* .get_rhs_packed_offset_ex = */ q4_k_gemm_adapter::get_rhs_packed_offset,
-            /* .run_kernel_ex         = */ q4_k_gemm_adapter::run,
+            /* .get_m_step            = */ &q4_k_get_m_step<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_n_step            = */ &q4_k_get_n_step<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_mr                = */ &q4_k_get_m_step<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_nr                = */ &q4_k_get_n_step<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_kr                = */ &q4_k_get_constant<4>,
+            /* .get_sr                = */ &q4_k_get_constant<2>,
+            /* .get_dst_offset        = */ &q4_k_get_dst_offset<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_lhs_offset_ex     = */ &q4_k_get_lhs_packed_offset<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .get_rhs_packed_offset_ex = */ &q4_k_get_rhs_packed_offset<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
+            /* .run_kernel_ex         = */ &q4_k_run_matmul<kai_matmul_clamp_f32_f16p4vsx2_qai4c32p16vsx4s1s0sf16_4vsx16vs_sme2_mopa>,
        },
        {
            /* .get_offset            = */ kai_get_lhs_offset_lhs_pack_f16pmrx2_f32_neon,
@@ -788,16 +756,16 @@ static ggml_kleidiai_kernels gemm_gemv_kernels_q4_k[] = {
        },
        /* SME2 GEMV */
        {
-            /* .get_m_step            = */ q4_k_gemv_adapter::get_m_step,
-            /* .get_n_step            = */ q4_k_gemv_adapter::get_n_step,
-            /* .get_mr                = */ q4_k_gemv_adapter::get_mr,
-            /* .get_nr                = */ q4_k_gemv_adapter::get_nr,
-            /* .get_kr                = */ q4_k_gemv_adapter::get_kr,
-            /* .get_sr                = */ q4_k_gemv_adapter::get_sr,
-            /* .get_dst_offset        = */ q4_k_gemv_adapter::get_dst_offset,
-            /* .get_lhs_offset_ex     = */ q4_k_gemv_adapter::get_lhs_packed_offset,
-            /* .get_rhs_packed_offset_ex = */ q4_k_gemv_adapter::get_rhs_packed_offset,
-            /* .run_kernel_ex         = */ q4_k_gemv_adapter::run,
+            /* .get_m_step            = */ &q4_k_get_m_step<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_n_step            = */ &q4_k_get_n_step<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_mr                = */ &q4_k_get_m_step<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_nr                = */ &q4_k_get_n_step<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_kr                = */ &q4_k_get_constant<4>,
+            /* .get_sr                = */ &q4_k_get_constant<2>,
+            /* .get_dst_offset        = */ &q4_k_get_dst_offset<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_lhs_offset_ex     = */ &q4_k_get_lhs_packed_offset<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .get_rhs_packed_offset_ex = */ &q4_k_get_rhs_packed_offset<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
+            /* .run_kernel_ex         = */ &q4_k_run_matmul<kai_matmul_clamp_f32_qsi8d32p1x4_qai4c32p16vsx4s1s0sf16_1x16vs_sme2_dot>,
        },
        {
            /* .get_offset            = */ q4_k_lhs_get_offset,
@@ -915,54 +883,6 @@ static ggml_kleidiai_kernels gemm_gemv_kernels_q8[] = {
        /* .rhs_type           = */ GGML_TYPE_Q8_0,
        /* .op_type            = */ GGML_TYPE_F32,
    },
-    {
-        /* SVE I8MM GEMM */
-        {
-            /* .get_m_step            = */ sve_q8_gemm_adapter::get_m_step,
-            /* .get_n_step            = */ sve_q8_gemm_adapter::get_n_step,
-            /* .get_mr                = */ sve_q8_gemm_adapter::get_mr,
-            /* .get_nr                = */ sve_q8_gemm_adapter::get_nr,
-            /* .get_kr                = */ sve_q8_gemm_adapter::get_kr,
-            /* .get_sr                = */ sve_q8_gemm_adapter::get_sr,
-            /* .get_dst_offset        = */ sve_q8_gemm_adapter::get_dst_offset,
-            /* .get_lhs_offset_ex     = */ sve_q8_gemm_adapter::get_lhs_packed_offset,
-            /* .get_rhs_packed_offset_ex = */ sve_q8_gemm_adapter::get_rhs_packed_offset,
-            /* .run_kernel_ex         = */ sve_q8_gemm_adapter::run,
-        },
-        /* .gemm_lhs_info = */ {
-            /* .get_offset            = */ kai_get_lhs_offset_lhs_quant_pack_qai8dxp_f32,
-            /* .get_packed_offset_ex  = */ &lhs_offs_fn5<kai_get_lhs_packed_offset_lhs_quant_pack_qai8dxp_f32>,
-            /* .packed_size_ex        = */ &lhs_ps_fn5<kai_get_lhs_packed_size_lhs_quant_pack_qai8dxp_f32>,
-            /* .pack_func_ex          = */ &lhs_pack_float_fn9_no_bl<kai_run_lhs_quant_pack_qai8dxp_f32>,
-        },
-        /* SVE DOT GEMV */
-        {
-            /* .get_m_step            = */ sve_q8_gemv_adapter::get_m_step,
-            /* .get_n_step            = */ sve_q8_gemv_adapter::get_n_step,
-            /* .get_mr                = */ sve_q8_gemv_adapter::get_mr,
-            /* .get_nr                = */ sve_q8_gemv_adapter::get_nr,
-            /* .get_kr                = */ sve_q8_gemv_adapter::get_kr,
-            /* .get_sr                = */ sve_q8_gemv_adapter::get_sr,
-            /* .get_dst_offset        = */ sve_q8_gemv_adapter::get_dst_offset,
-            /* .get_lhs_offset_ex     = */ sve_q8_gemv_adapter::get_lhs_packed_offset,
-            /* .get_rhs_packed_offset_ex = */ sve_q8_gemv_adapter::get_rhs_packed_offset,
-            /* .run_kernel_ex         = */ sve_q8_gemv_adapter::run,
-        },
-        /* .gemv_lhs_info = */ {
-            /* .get_offset            = */ kai_get_lhs_offset_lhs_quant_pack_qai8dxp_f32,
-            /* .get_packed_offset_ex  = */ &lhs_offs_fn5<kai_get_lhs_packed_offset_lhs_quant_pack_qai8dxp_f32>,
-            /* .packed_size_ex        = */ &lhs_ps_fn5<kai_get_lhs_packed_size_lhs_quant_pack_qai8dxp_f32>,
-            /* .pack_func_ex          = */ &lhs_pack_float_fn9_no_bl<kai_run_lhs_quant_pack_qai8dxp_f32>,
-        },
-        /* .rhs_info = */ {
-            /* .packed_size_ex        = */ &rhs_ps_fn5<kai_get_rhs_packed_size_rhs_pack_nxk_qsi8cxp_qsi8cx_neon>,
-            /* .pack_func_ex          = */ &rhs_pack_scale_fn12<kai_run_rhs_pack_nxk_qsi8cxp_qsi8cx_neon>,
-        },
-        /* .required_cpu       = */ CPU_FEATURE_SVE | CPU_FEATURE_SVEI8MM | CPU_FEATURE_DOTPROD,
-        /* .lhs_type           = */ GGML_TYPE_F32,
-        /* .rhs_type           = */ GGML_TYPE_Q8_0,
-        /* .op_type            = */ GGML_TYPE_F32,
-    },
    {
        /* I8MM GEMM */
        {
@@ -13,8 +13,7 @@ enum cpu_feature {
    CPU_FEATURE_SVE     = 4,
    CPU_FEATURE_SME     = 8,
    CPU_FEATURE_SME2    = 16,
-    CPU_FEATURE_FP16    = 32,
-    CPU_FEATURE_SVEI8MM = 64
+    CPU_FEATURE_FP16    = 32
 };

 inline cpu_feature& operator|=(cpu_feature& lhs, cpu_feature rhs) {
@@ -319,10 +319,8 @@ static void init_kleidiai_context(void) {

        size_t detected_smcus = 0;

-        // The fixed-width SVE kernels require 32-byte vectors, including the Q8 shared RHS layout.
        ctx.features  = (runtime_feat.has_dotprod  ? CPU_FEATURE_DOTPROD : CPU_FEATURE_NONE) |
                        (runtime_feat.has_i8mm     ? CPU_FEATURE_I8MM    : CPU_FEATURE_NONE) |
-                        (runtime_feat.has_svei8mm  ? CPU_FEATURE_SVEI8MM : CPU_FEATURE_NONE) |
                        (runtime_feat.has_fp16     ? CPU_FEATURE_FP16    : CPU_FEATURE_NONE) |
                        (runtime_feat.sve_cnt == QK8_0 ? CPU_FEATURE_SVE : CPU_FEATURE_NONE);

@@ -1618,11 +1616,6 @@ public:
            const size_t kr = kernel->get_kr();
            const size_t sr = kernel->get_sr();
            const ggml_type rhs_type = kernels->rhs_type;
-            if (rhs_type == GGML_TYPE_Q8_0 && (kernels->required_cpu & CPU_FEATURE_SVE)) {
-                GGML_ASSERT(nr == kernels->gemv.get_nr());
-                GGML_ASSERT(kr == kernels->gemv.get_kr());
-                GGML_ASSERT(sr == kernels->gemv.get_sr());
-            }
            const size_t block_len = rhs_type == GGML_TYPE_Q8_0 ? QK8_0 :
                                     rhs_type == GGML_TYPE_Q4_0 ? QK4_0 :
                                     rhs_type == GGML_TYPE_Q4_K ? GGML_KLEIDIAI_Q4_K_BL :
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Aleksander Grygier	e285cfe2d8	ui: spin the compaction icon while the summary runs The divider reported the phase in its title while the icon still showed a finished block, so a run in progress and a summary that had landed looked the same in the one place the eye reads first. The icon spins while the summary is running and folds back once it is in.	2026-10-21 10:38:23 +02:00
Aleksander Grygier	6f6689fe0b	ui: open a compaction summary at the top of the screen Toggling a summary left the reader wherever they were, so a conversation that had just compacted showed the divider at the bottom of the screen and read down from there. A toggle now scrolls the page once and lands the divider a breath under the tabs, both ways: the summary is read from its beginning, and closing it returns to the same boundary instead of wherever the page happens to come to rest. The scroll waits for the body to reach its height, because before that the page is too short to scroll the divider up there and the call would clamp. That wait is the shared collapse duration rather than a transition event: an engine without `interpolate-size` animates the region's grid row instead of its height, so a listener on `height` would never fire there and the page would not move at all.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	883f7387bd	ui: expand collapsible bodies with a height transition Expanding a block snapped it open: the body was mounted the moment `open` flipped, so a tall block moved the conversation around it in one frame. The body now goes through the same region the models table uses. The height interpolates and the visibility follows it, so a body is unmounted once the collapse has finished rather than the moment it starts - a closed block still costs nothing while tokens stream past it.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	2894120ee1	ui: follow the bottom, otherwise leave the view alone The screen scrolled itself with a controller of observers, intervals, direction heuristics and landing timers, and still managed to move the view while the reader was reading: any mutation that grew the branch was treated as a reason to keep following. The page is the scroller, so it follows when the reader sits at the bottom, and stays where it is otherwise. A resize observer plus native scroll anchoring does the rest, and `pointerdown` buys the reader a moment of grace to be caught mid-drag.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	857354c064	ui: hand the compact action to the context gauge Compacting is a decision about the context, so the action belongs with the numbers that show the context filling up rather than inside the form: it sits at the bottom of both the hover card and the drawer, under the breakdown it acts on, quiet enough to stay a footnote and disabled - with the row's own spinner - while a summary is already running.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	3b1e02fd69	ui: keep the summarized turns quiet while a compaction runs The turns being summarized used to keep their own "Processing..." indicator and live reading stats while the summary streamed, so the conversation said the same thing twice in two places. The row reports the run, so the turns it replaces stay silent until it is done.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	accaf63d41	ui: report compaction progress on the row The row owns the run, so the row says how it is going: the prompt progress in the header while the context is being read, the phase once the summary starts streaming, and the token count once it is in. Before the first summary token there is nothing to collapse, so the header stands on its own instead of offering an empty block.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	584326fcec	ui: give the compaction row its own summary UI The summary was a line of text in a tool-call-shaped collapsible, which is how a compaction reads as a side note rather than as the thing the rest of the conversation now stands on. The row is its own block: a divider that names the boundary, and the summary inside it with the model that wrote it, its statistics, and the same actions an assistant turn has - copy, edit, fork, regenerate and delete all apply to the summary, and it collapses to the divider once a summary is in, because the shrunk context is the headline and the prose is the detail. The divider is a variant of the shared collapsible block, which also needed the toggle and the title split into snippets so a caller can lay them out without copying the markup. The action icons no longer take a role: they render what a message offers, and a compaction offers the same as any other stored message.	2026-10-21 10:27:29 +02:00
Aleksander Grygier	afe06daa52	ui: run every compaction summary through one path Three callers needed the same thing - the automatic pre-encode pass, the action in the context gauge, and re-running a summary from the block - and each of them had grown its own copy of the request, the streaming and the storing. There is one path now: a summarization owns a node, streams into it, and reports itself through a ledger the rest of the app reads (`isCompacting`, which node belongs to which conversation). Re-running a summary is the same call on an existing node, so the two entry points cannot drift apart again.	2026-10-21 10:27:29 +02:00
Pascal	815edb76ad	ui: follow the moved models selector in the compaction model field The models manager moved ModelsSelectorList under ModelsSelector and the model option helpers to /utils, where groupModelOptions takes the options alone. Drop the stale copy the rebase kept and read the helpers from their new home.	2026-10-21 09:45:43 +02:00
Pascal	6e24c83b2f	ui: import the model option helpers from models/utils	2026-10-21 09:39:23 +02:00
Pascal	e138b82720	ui: fix the model option helpers import path The compaction model selector pulls filterModelOptions and groupModelOptions from the models component folder, where they no longer live, so the build fails to resolve the module. Point the import at the navigation folder, which now owns them. Missed during the rebase: the helpers moved in a commit that does not touch any compaction file, so nothing conflicted.	2026-10-21 09:39:23 +02:00
Pascal	f9fc748df4	ui: align the compaction model field with project conventions	2026-10-21 09:39:23 +02:00
Pascal	cddfa89091	ui: allow a dedicated model for conversation compaction An opt-in setting reveals a model selector whose options come from the live router model list. The chosen model writes the summary and is loaded on demand when it is not resident, while the default path keeps resolving to the conversation's own model.	2026-10-21 09:39:23 +02:00
Pascal	a8187b8b85	chore: format	2026-10-21 09:39:10 +02:00
Pascal	ade61a6d7a	ui: pre-fill KV cache and refresh context gauge after compaction Once the summary is stored, pre-encode the compacted branch and stamp the returned prompt timings on the compaction node: re-encoding the reduced prefix is part of the compaction itself, so the server KV cache and the context gauge reflect the summary right away. The summary stream is registered under a prefixed identity (compact::convId), so the visibility kick and byte-offset resume protect it through tab switches while the reload probe never reattaches it as a chat turn. The stamped node becomes the shared context truth for the gauge and the processing state restore, and the stale live state that kept the gauge full is cleared.	2026-10-21 09:36:46 +02:00
Pascal	0543f7ba85	ui: add conversation compaction Add a compaction action that asks the model to summarize the conversation so far and stores the summary in a new compaction node appended to the branch. Later requests slice the branch at the last compaction node, keeping the system prefix and continuing from the summary instead of the full history, which frees up context while the complete history stays visible and navigable in the UI. The slice is applied once at the request funnel, covering the plain send, the agentic flow and the KV cache pre-encode, plus the continue path that bypasses it. The summary request is a one-shot outside the SSE replay machinery, streamed live into the node with a full rollback on abort, error or empty result. The action is exposed as a /compact slash command and a Compact button in the context gauge popup, with an overridable summarization prompt in the settings.	2026-10-21 09:36:45 +02:00