ggml-org/llama.cpp

Pull Requests Commits

0cc4m committed 2 days ago

7d7437a3

vulkan: tune GDN shader

0cc4m committed 2 days ago

b0979ca5

polished Readme and llama-bench (#28968)

truecoder34 committed 2 days ago

Verified 965f8979

ggml-cpu: tiled mul_mat for k-quants (#27851)

jbooth committed 2 days ago

Verified d834d44e

opencl: add A8 Q8_0 non-MoE dp4a binary kernel (#29439)

shaofeiqi committed 2 days ago

Verified 9f70b2ce

hexagon: find software divide calls using binary inspection tool (#29449)

trivikram-reddy1 committed 2 days ago

Verified 4e748117

vendor : update cpp-httplib to 0.58.0 (#29407)

cabelo committed 2 days ago

Verified 171e8846

common,rpc : simplify fs_create_directory_with_parents() (#29432)

angt committed 3 days ago

Verified 4b1a27fa

mtmd: fix mel preprocessor in LFM2 audio (#29403)

ykhrustalev committed 3 days ago

Verified fcc89154

opencl: add bin kernel `kernel_gemm_noshuffle_q5_k_f32_32b_trans_ila_a8_bin`, `kernel_gemm_noshuffle_q5_k_q8_1_dp4a_ila_a8_bin` (#29401)

shaofeiqi committed 3 days ago

Verified a25c9865

Fixing the vulkan build issue of legacy GLSLC version that has no cooperativeMatrix API support (https://github.com/ggml-org/llama.cpp/issues/29373) (#29409)

sliu39 committed 3 days ago

Verified e85e15cf

gguf-py : ByteLevel processing defaults bos/eos to False (#29422)

CISC committed 3 days ago

Verified b248f4a3

gguf-py : TemplateProcessing has final word on add_special_token (#29417)

CISC committed 3 days ago

Verified d81aef19

common : extract shared unicode path/string helpers (#29415)

angt committed 3 days ago

Verified 27b20ba8

metal: FWHT kernels for block widths above 512 (#29095)

bri-prism committed 3 days ago

Verified e351231c

metal : split fa kernels into per-dtype libraries (#29329)

ggerganov committed 3 days ago

Verified 5a75f14c

llama : add `llama_prec_policy` + model-driven W4A4 path (#24364)

ynankani committed 3 days ago

Verified e9f824d8

HIP: bump HIP_VERSION requried for fp8 to avoid missing __hip_fp8_e4m3 support in 6.2 (#29231)

IMbackK committed 3 days ago

Verified d028c697

rpc: include nb in the get_alloc_size cache key and floor the result at ggml_nbytes (#29283)

Jesssullivan committed 3 days ago

Verified 66963a8b

[SYCL] support sparse FA (#28796)

arthw committed 3 days ago

Verified cd74ef62

musa: fix PH1 (MTT S5000) operator failures and build issues (#29193)

yeahdongcn committed 3 days ago

Verified f9af9be2

CUDA: fuse RMS_NORM + SCALE into one kernel (#29393)

InflexCZE committed 3 days ago

Verified 1ab7e5ad

llama : fix tensor split for fused qkv with uneven K/V head sizes (#29294)

am17an committed 3 days ago

Verified f805c57a

hexagon: add q5_k quant type support (#29123)

jhen0409 committed 3 days ago

Verified 4de09265

hexagon: use DMA for contiguous dim1 CONCAT (#29404)

kurquhar committed 3 days ago

Verified ed319feb

metal : fix graph capture and handle empty graphs (#29390)

ggerganov committed 4 days ago

Verified 84e76d8a

metal : optimize sparse FA + clean-up (#29377)

ggerganov committed 4 days ago

Verified cdc06426

sync : ggml (#29396)

ggerganov committed 4 days ago

Verified bced4595

hexagon: handle multi-sequence in concat_2d (#29344)

jhen0409 committed 4 days ago

Verified a02c7f58

llama-grammar: fix numeric truncation for token_id parsing (#29382)

apach301 committed 4 days ago

Verified 5cf3a352

Older