upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-23 11:10:55 -05:00
Code Issues Packages Projects Releases Wiki Activity

spec : add DFlash support (#22105)

Browse Source 
* spec: add DFlash v2 support

* dflash: support sliding window attention per layer_types

* docs: add dflash section

---------

Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>
This commit is contained in:
Ruixiang Wang
2026-06-28 15:01:34 +02:00
committed by GitHub
parent c1a1c8ee94
commit d1b34251bc
14 changed files with 712 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
gguf-py/gguf/constants.py
View File

@@ -517,6 +517,7 @@ class MODEL_ARCH(IntEnum):
PANGU_EMBED = auto()
MISTRAL3 = auto()
EAGLE3 = auto()
DFLASH = auto()
MISTRAL4 = auto()
PADDLEOCR = auto()
MIMO2 = auto()
@@ -1074,6 +1075,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.PANGU_EMBED: "pangu-embedded",
MODEL_ARCH.MISTRAL3: "mistral3",
MODEL_ARCH.EAGLE3: "eagle3",
MODEL_ARCH.DFLASH: "dflash",
MODEL_ARCH.MISTRAL4: "mistral4",
MODEL_ARCH.PADDLEOCR: "paddleocr",
MODEL_ARCH.MIMO2: "mimo2",
@@ -4086,6 +4088,22 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FC,
MODEL_TENSOR.D2T,
],
MODEL_ARCH.DFLASH: [
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_Q_NORM,
MODEL_TENSOR.ATTN_K_NORM,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FC,
MODEL_TENSOR.ENC_OUTPUT_NORM,
],
MODEL_ARCH.MISTRAL4: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,