ochafik/llama.cpp

Pull Requests Commits

server : return unparseable output as content instead of erroring

ochafik committed 15 days ago

9ea3f944

grammar : only arm a lazy grammar if the trigger is followed by parseable output

ochafik committed 15 days ago

1a909153

RPC: use RDMA completion channel to not spin (#29440)

am17an committed 15 days ago

Verified d7fb90e8

ci : enable GGML_SCHED_DEBUG_REALLOC=1 for ctest workflows (#29514)

ggerganov committed 15 days ago

Verified 7fb2b082

llama-bench : fix OOB access of hf_file (#29515)

angt committed 15 days ago

Verified 187664b5

hrm : fix layer placement of `z_l_init` weight (#29512)

ggerganov committed 15 days ago

Verified 85ca3b52

hexagon: support tiled Q4_0 and Q8_0 GET_ROWS (#29511)

kurquhar committed 15 days ago

Verified 7ac59a6e

hexagon: support for backend sampler (#29502)

max-krasnyansky committed 15 days ago

Verified 2b129ccf

cuda: support Nemotron 3 Puzzle state size 96 for ssm scan (#28717)

anavp-nvidia committed 16 days ago

Verified 95887577

musa: build the docker images from the PH1 MUSA SDK image (#29481)

yeahdongcn committed 16 days ago

Verified 694ec235

cuda: add F16 input to the FWHT (#29096)

bri-prism committed 16 days ago

Verified 6f856c70

server : fix wake_fd warning on Windows (#29479)

angt committed 16 days ago

Verified fcb3074f

Revert "Change max context length for auto-fitting with unified KV (#28849)" (#29437)

gaugarg-nv committed 16 days ago

Verified 2145525a

jinja : implement sameas test (#29448)

CISC committed 16 days ago

Verified 81bc6b83

jinja : fix compile error (#29468)

CISC committed 16 days ago

Verified 86a24a18

llama : fix K/V and recurrent state cleanup after failed restores (#27530)

CHIPMUNK-T0T committed 16 days ago

Verified 08618ff8

jinja : support noncall test statements with arg (#29443)

CISC committed 16 days ago

Verified a1de614b

polished Readme and llama-bench (#28968)

truecoder34 committed 16 days ago

Verified 965f8979

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

jbooth committed 16 days ago

Verified d834d44e

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

shaofeiqi committed 16 days ago

Verified 9f70b2ce

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

trivikram-reddy1 committed 16 days ago

Verified 4e748117

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

cabelo committed 16 days ago

Verified 171e8846

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

angt committed 17 days ago

Verified 4b1a27fa

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

ykhrustalev committed 17 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 17 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 17 days ago

Verified e85e15cf

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

CISC committed 17 days ago

Verified b248f4a3

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

CISC committed 17 days ago

Verified d81aef19

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

angt committed 17 days ago

Verified 27b20ba8

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

bri-prism committed 17 days ago

Verified e351231c

Older