ggml-org/ggml

Pull Requests Commits

sync : llama.cpp

ggerganov committed 15 hours ago

58c38058

metal : add missing mm-id specializations for q1_0 (llama/21662)

ggerganov committed 15 hours ago

460533c1

sycl : add flash-attn support for head size 512 (llama/21654)

qnixsynapse committed 15 hours ago

b5cae268

vulkan: unify type macros to use Vx instead of _VECx (llama/21605)

0cc4m committed 15 hours ago

16a9e2c1

CUDA: also store `node->src->data` ptrs for equality check (llama/21635)

am17an committed 15 hours ago

e98b2f42

fix: free ctx_copy in ggml_opt_free to plug per-training-session leak (llama/21592)

RealOrko committed 15 hours ago

e13cec10

webgpu : Query for adapter support when registering WebGPU backend (llama/21579)

reeselevine committed 15 hours ago

5eeebddf

metal: Q1_0 backend (llama/21528)

khosravipasha committed 15 hours ago

91dc6fc4

CUDA: make cuda graphs props check faster (llama/21472)

am17an committed 15 hours ago

197e255a

ggml-cuda: ds_read_b128 for q4_0 and q4_1 mmq kernels (llama/21168)

iacopPBK committed 15 hours ago

327b6600

ggml-webgpu: parameterize submission size and add iOS specific limits (llama/21533)

reeselevine committed 15 hours ago

784d93ba

CUDA: check for buffer overlap before fusing (llama/21566)

am17an committed 15 hours ago

3dba4034

ggml : deprecate GGML_OP_ADD1 (llama/21363)

ggerganov committed 15 hours ago

b75eb9a1

ggml: Vulkan build, Linux -- output error string for errno on fork failure (#20868) (llama/20904)

tomoverlund committed 15 hours ago

484f2d87

vulkan: add FA dequant for q4_1, q5_0, q5_1, iq4_nl (llama/21029)

mkoker committed 15 hours ago

c1fd77ef

ggml-cuda : fix CDNA2 compute capability constant for gfx90a (MI210) (llama/21519)

aviallon committed 15 hours ago

f162d653

Add Q8_0 reorder optimization (~3x tg speedup on Intel Arc) (llama/21527)

PMZFX committed 15 hours ago

a50a58ac

ggml-webgpu: Add the support of `MUL_MAT_ID` (llama/21147)

yomaytk committed 15 hours ago

89613f18

ggml: add Q1_0 1-bit quantization support (CPU) (llama/21273)

khosravipasha committed 15 hours ago

b5dafbb0

Write an optimized flash_attn_stream_k_fixup kernel (llama/21159)

gaugarg-nv committed 15 hours ago

30ad54ee

sycl : handle other FA case (llama/21377)

arthw committed 15 hours ago

d36caad9

hexagon: slight optimization for argosrt output init (llama/21463)

YardenTal44 committed 15 hours ago

1b7b7b13

ggml-webgpu: move from parameter buffer pool to single buffer with offsets (llama/21278)

reeselevine committed 15 hours ago

67767f79

ggml-zendnn : add MUL_MAT_ID op support for MoE models (llama/21315)

z-vishal committed 15 hours ago

741295d7

rpc : reuse compute graph buffers (llama/21299)

rgerganov committed 15 hours ago

70bf4fc0

ggml-webgpu: add vectorized flash attention (llama/20709)

ArberSephirotheca committed 15 hours ago

3ccdf784

sycl : fix llama_kv_cache hang when kv_cache is huge: 5GB (llama/21283)

arthw committed 15 hours ago

34834a85

ggml : bump version to 0.9.11 (#1456)

ggerganov committed 7 days ago

Verified 49f84a92

sync : llama.cpp

ggerganov committed 7 days ago

c3d71fa1

hexagon : add cumsum op support (llama/21246)

tboinovski1 committed 7 days ago

b26daea9

Older