Commit Graph

  • e9dbd0c18a ggml-webgpu: FlashAttention refactor + standardize quantization support (llama/23834) Reese Levine 2026-06-03 22:05:04 -07:00
  • a1a3186887 ggml-cpu: extend RVV quantization vec dot to higher VLENs (llama/22754) rehan-10xengineer 2026-06-04 10:03:40 +05:00
  • 00a9728de3 Avoid PDL race conditions by disabling __restrict__ when PDL is used (llama/24030) Andreas Kieslinger 2026-06-03 13:56:42 +02:00
  • 750fa4ca35 ggml-cpu: use runtime SVE width in FWHT (llama/24059) Charles Xu 2026-06-03 12:45:10 +02:00
  • d5a49ebec8 cuda: reserve space for quantize kv-cache at startup (llama/23907) Aman Gupta 2026-06-03 18:39:59 +08:00
  • f110ff540c opencl: use flat variants of q4_K and q6_K gemv for very large M (llama/24006) lhez 2026-06-02 14:16:17 -07:00
  • d31cb20b25 hexagon: profiler output fix and script updates (llama/24042) Max Krasnyansky 2026-06-02 14:08:29 -07:00
  • 8d61a9edf0 hexagon: MUL_MAT, MUL_MAT_ID, FLASH_ATTN and GDN cleanup and optimizations for latest models (llama/23989) Max Krasnyansky 2026-06-01 23:40:08 -07:00
  • 754247f28b hexagon: add gelu_quick (llama/24007) Todor Boinovski 2026-06-01 23:19:07 -07:00
  • 79223704a1 clean up unused variables warnings (llama/23975) Anav Prasad 2026-06-01 19:38:37 -07:00
  • 9a0265d13b opencl: fix compiler warnings for non-adreno path (llama/23922) lhez 2026-06-01 19:15:09 -07:00
  • db2a39507c revert to using global_invocation_id for cpy shader (llama/23955) Masashi Yoshimura 2026-06-02 08:59:06 +09:00
  • e728bae159 opencl: add basic support for q5_0 and q5_1 (llama/23548) shaofeiqi 2026-06-01 10:06:50 -07:00
  • 050b8567a0 metal: template GLU kernels to support f16/f32 (llama/23882) Shrivas Shankar 2026-06-01 07:40:28 -05:00
  • 71d80aa49e vulkan: don't hold the device mutex while compiling pipelines (llama/23641) Jeff Bolz 2026-06-01 07:04:01 -05:00
  • c471bcce1b vulkan: reduce host memory lock contention (llama/23376) Winston Ma 2026-06-01 20:03:32 +08:00
  • e815b264eb TP: quantized KV cache support (llama/23792) Johannes Gäßler 2026-06-01 12:30:10 +02:00
  • 982533fc0c vulkan: Block-load Q3_K/Q6_K block data and subtract on 32b ints (llama/23056) Matt Corallo 2026-06-01 09:46:48 +00:00
  • aea93ada61 vulkan: Removed unused functions (llama/23175) Winston Ma 2026-06-01 17:46:23 +08:00
  • ec0c661950 Support Q4_1, Q5_0, Q5_1 in Flash-attention (llama/23812) Neo Zhang 2026-06-01 14:53:53 +08:00
  • 20323e48c4 Add more types in GET_ROWS OP (llama/23710) Neo Zhang 2026-06-01 14:53:04 +08:00
  • 687fbcb149 sycl : Optimize Q3_K mul_mat by reorder (llama/23725) Neo Zhang 2026-06-01 14:50:55 +08:00
  • 1c0d1f0f7c opencl: support bf16 by converting to f16 (llama/23839) lhez 2026-05-30 10:17:47 -07:00
  • bf74b557d2 metal : restore im2col implementation for large kernels (llama/23901) Georgi Gerganov 2026-05-30 15:26:13 +03:00
  • 64b0d6b7fc ggml : add some lsx support (llama/23798) Jinyang He 2026-05-30 16:53:26 +08:00
  • 4317ddbe2b vulkan: add Flash Attention support for BFloat16 KV cache (llama/23420) Ruben Ortlam 2026-05-30 10:39:31 +02:00
  • 9147a9676b ggml-webgpu: Check earlier for WebGPU required features (llama/23879) Reese Levine 2026-05-29 14:16:05 -07:00
  • acd91d2c38 ggml-webgpu: add q4_0/q8_0 SET_ROWS (llama/23760) Reese Levine 2026-05-29 14:14:11 -07:00
  • f7aad4ed7e CUDA: Check PTX version on host side to guard PDL dispatch (llama/23530) Oliver Simons 2026-05-29 12:28:18 +02:00
  • c50e951afd model : support for DeepseekV32ForCausalLM with generic DeepSeek Sparse Attention (DSA) implementation (llama/23346) fairydreaming 2026-05-29 10:15:17 +02:00
  • e1da83d773 ci : add ccache to build-sycl [no ci] (#3859) Daniel Bevenius 2026-06-08 07:27:12 +02:00
  • a8ec021f27 ci : add HF_TOKEN to docker.yml workflow [no ci] (#3861) Daniel Bevenius 2026-06-06 18:34:40 +02:00
  • 574fc0da69 ci : add ccache to quantize, vad, and wasm jobs (#3860) Daniel Bevenius 2026-06-06 05:40:58 +02:00
  • 99613cb720 ci: build-windows action slimming (#3858) Daniel Bevenius 2026-06-04 16:27:58 +02:00
  • ad17783d34 ci : use emscripten-core and pin version (#3857) Daniel Bevenius 2026-06-04 14:25:15 +02:00
  • 7ecb08f263 ci : pin github actions to commit SHAs (#3856) Daniel Bevenius 2026-06-04 11:38:46 +02:00
  • 9302c060f0 ci : use ccache instead of sccache for windows-cublas [no ci] (#3855) Daniel Bevenius 2026-06-04 11:37:22 +02:00
  • 12d1828837 ci : only publish/push docker images daily (#3854) Daniel Bevenius 2026-06-04 10:30:48 +02:00
  • 02d5316af5 ci : refactor + optimize (#3847) Georgi Gerganov 2026-06-04 09:35:58 +03:00
  • 610e664ba7 whisper : catch C++ exceptions in whisper_init_with_params_no_state (#3831) danscMax 2026-06-02 14:25:29 +03:00
  • e5d4412578 server : merge split utf-8 token text in verbose json (#3850) Noah Lyons 2026-06-02 07:10:27 -04:00
  • ef24de1e58 cmake : do not assume /usr/lib library installation. (#3693) Patrice Levesque 2026-06-02 03:22:16 -04:00
  • 23ee03506a release : v1.8.6 v1.8.6 Georgi Gerganov 2026-06-01 14:56:20 +03:00
  • 0dff27498f ci : fix path to whisper.h in examples.yml [no ci] (#3842) Daniel Bevenius 2026-06-01 07:20:19 +02:00
  • fe69461618 ci : fix self-hosted paths to mnt Georgi Gerganov 2026-05-31 16:06:32 +03:00
  • 099af1c67d pi : add config Georgi Gerganov 2026-05-31 16:04:12 +03:00
  • 2e045a967b ci : remove obsolete self-hosted label Georgi Gerganov 2026-05-31 15:45:44 +03:00
  • 6c343e7a4e common : pass sample rate to ffmpeg_decode_audio() Georgi Gerganov 2026-05-31 15:48:05 +03:00
  • d418d645a0 common : pass sample rate to ffmpeg_decode_audio() gg/ci-remove-label Georgi Gerganov 2026-05-31 15:48:05 +03:00
  • 8d7d544562 ci : remove obsolete self-hosted label Georgi Gerganov 2026-05-31 15:45:44 +03:00
  • f39cc71282 common : re-implement ffmpeg-transcode.cpp + clarify ffmpeg usage (#3846) Georgi Gerganov 2026-05-31 15:44:07 +03:00
  • f24588a272 sync : ggml v1.8.5 Georgi Gerganov 2026-05-29 09:46:42 +03:00
  • 92fc3f2a58 ggml : bump version to 0.13.1 (ggml/1523) Georgi Gerganov 2026-05-29 09:46:12 +03:00
  • 5828fba79f talk-llama : sync llama.cpp Georgi Gerganov 2026-05-29 09:44:28 +03:00
  • cc65eb1816 sync : ggml Georgi Gerganov 2026-05-29 09:43:15 +03:00
  • e90501e179 cuda : disables launch_fattn PDL enrollment due to compiler bug (llama/23825) Andreas Kieslinger 2026-05-29 06:46:10 +02:00
  • f1b687da28 meta : Add missing buffer set in allreduce fallback !COMPUTE clear (llama/23480) Matt Corallo 2026-05-29 03:30:24 +00:00
  • 442be1789d hexagon: basic/generic op fusion support and RMS_NORM+MUL fusion (llama/23835) Max Krasnyansky 2026-05-28 14:05:54 -07:00
  • 94922ce12c opencl: move backend info printing into its own function (llama/23702) lhez 2026-05-28 11:05:42 -07:00
  • e1faa7cb4d ggml: auto apply iGPU flag CUDA/HIP if integrated device (llama/23007) fl0rianr 2026-05-28 15:01:14 +02:00
  • 4e8af441e5 mmvq Optim: add MMVQ_PARAMETERS_TURING(mmvq_parameter_table_id) for … (#23729) redfox 2026-05-28 20:51:14 +08:00
  • 04795e6272 CUDA: route batch>=4 quantized matmul to MMQ on AMD MFMA hardware (llama/23227) Jaden_Mach 2026-05-28 08:50:25 -04:00
  • 1b241b879c hexagon: minor refresh for HMX FA and MM (llama/23796) Max Krasnyansky 2026-05-28 04:49:11 -07:00
  • b896e91f18 vulkan: fast path for walsh-hadamard transform (llama/23687) Jeff Bolz 2026-05-28 06:18:43 -05:00
  • 816c3029bc vulkan: fix wrong index variable in inner loop (llama/23665) Winston Ma 2026-05-28 18:48:34 +08:00
  • 5db94bac04 vulkan: Fix memory logger unsafe iterator access (llama/23667) Winston Ma 2026-05-28 18:46:07 +08:00
  • 60e420ff6a cuda : fix KQ mask offset integer overflow in fattn MMA kernel (llama/23610) fairydreaming 2026-05-28 10:55:42 +02:00
  • 8e40325876 ggml: fixed Arm SVE usage bug in vec.h, vec.cpp (llama/22841) Martin Klacer 2026-05-28 08:04:21 +01:00
  • d284e1c3aa Hexagon: OP_GATED_DELTA_NET K>1 support (llama/23531) ymcki 2026-05-28 14:05:25 +08:00
  • 7e843a80e1 opencl: OP_GATED_DELTA_NET (llama/23312) ymcki 2026-05-28 12:23:21 +08:00
  • 8c8f213dac ggml-webgpu: remove legacy constants (llama/23672) Reese Levine 2026-05-27 14:22:33 -07:00
  • 3bbe93378c hexagon: add support for Q4_1 in MUL_MAT and MUL_MAT_ID (llama/23647) Max Krasnyansky 2026-05-27 10:46:11 -07:00
  • a52bd385d6 ggml-webgpu: Fix how to dispatch WG to some ops (llama/23750) Masashi Yoshimura 2026-05-28 01:48:12 +09:00
  • 8bce478ee8 vulkan: Switch MUL_MAT_VEC to 4 K per iteration for F16/32 (llama/22887) Matt Corallo 2026-05-27 15:19:23 +00:00
  • 1b590bbb9a vulkan: use GL_NV_cooperative_matrix_decode_vector for faster matmul (llama/23541) Jeff Bolz 2026-05-27 10:18:28 -05:00
  • c5cde8c717 vulkan: add REPEAT op support for f16 to f16. (llama/23298) l8bloom 2026-05-27 16:59:08 +02:00
  • 98c6722fec CUDA: restrict PDL to CTK >= 12.3 due to MSVC issues (llama/23742) Oliver Simons 2026-05-27 14:21:04 +02:00
  • 80e87ec453 vulkan: avoid preferring transfer queue on AMD UMA devices (llama/22455) Winston Ma 2026-05-27 17:48:40 +08:00
  • 6a249cd640 ggml-zendnn : fixed naming of matmul function (llama/20964) Vladislav 2026-05-27 01:59:35 +03:00
  • a0efd13f0f vulkan: optimize conv2d and implement coopmat1 support (llama/22620) Jeff Bolz 2026-05-26 08:48:05 -05:00
  • f8df28d331 hexagon: add support for CONCAT op (llama/23648) Max Krasnyansky 2026-05-26 06:20:05 -07:00
  • 049f0af339 SYCL: implement ggml_sycl_pool_vmm (llama/22862) Alexey Kopytko 2026-05-26 13:59:00 +09:00
  • 00a5110b19 ggml-webgpu: Add MMVQ path for Q4/Q8/Q2_K/Q4_K and clean up legacy MUL_MAT pipeline (llama/23594) Masashi Yoshimura 2026-05-26 12:42:49 +09:00
  • bc77933c2d Check batch_compute_passes before sending passes when not doing GPU profiling (llama/23457) Nikhil Jain 2026-05-25 20:32:49 -07:00
  • 2307712d32 CUDA: missing PDL sync for FWHT, better fallback (llama/23690) Johannes Gäßler 2026-05-26 05:05:51 +02:00
  • 1c477d4056 metal : add apple device id (llama/23566) forforever73 2026-05-26 02:05:16 +08:00
  • 205ee5a189 CUDA: add fast walsh-hadamard transform (llama/23615) Aman Gupta 2026-05-25 21:12:10 +08:00
  • c932729a30 ci : add ignore for bindings/{ruby, go} in build.yml [no ci] (#3837) Daniel Bevenius 2026-05-28 18:06:04 +02:00
  • e47a3eeb04 ci : fix include paths for bindings-go job [no ci] (#3835) Daniel Bevenius 2026-05-28 14:53:34 +02:00
  • f41562bdd6 ci : add on push/pull_request paths ruby job (#3833) Daniel Bevenius 2026-05-28 14:41:48 +02:00
  • 9186e2453b ci : renable arm64 docker builds (#3832) Daniel Bevenius 2026-05-28 12:09:13 +02:00
  • f6e617bab7 ci : set GGML_NATIVE=OFF for bindings-java (#3830) Daniel Bevenius 2026-05-28 07:21:25 +02:00
  • 6dcdd65364 ci : only run docker jobs when pushed to master [no ci] (#3828) Daniel Bevenius 2026-05-27 08:46:23 +02:00
  • ee540bf0be docs : add AGENTS.md and CONTRIBUTING.md [no ci] (#3826) Daniel Bevenius 2026-05-27 06:22:38 +02:00
  • 27101c01dc cli : merge tokens split across UTF-8 boundaries in JSON output (#3751) texasich 2026-05-25 23:23:41 -05:00
  • e0fd1f6787 release : v1.8.5 Georgi Gerganov 2026-05-25 13:06:33 +03:00
  • c245b3ec23 benches : update Georgi Gerganov 2026-05-25 13:05:30 +03:00
  • f14ae77f40 sync : ggml Georgi Gerganov 2026-05-25 12:44:07 +03:00
  • 1cf8e3a903 ggml : bump version to 0.13.0 (ggml/1510) Georgi Gerganov 2026-05-25 12:40:17 +03:00
  • bcff515150 TP: fix ggml context size calculation (llama/22616) Johannes Gäßler 2026-05-25 11:37:25 +02:00