ggerganov/llama.cpp

Pull Requests Commits

cont : reduce paths

CISC committed 5 days ago

Verified 15324f90

cont : split server.yml

ggerganov committed 5 days ago

Verified 3fec0e1b

ci : split build.yml

ggerganov committed 5 days ago

Verified 45a8ab2e

ggml : guard against sumq2 being 0 in IQ4_NL (#20460)

bartowski1182 committed 5 days ago

Verified b9da4444

cuda : add RDNA4-specific MMVQ parameter table for bs=1 decode (#19478)

JoursBleu committed 5 days ago

Verified 617db241

vulkan: use graphics queue on AMD (#20551)

0cc4m committed 5 days ago

Verified 1a3d8edb

kv-cache : fix reading llama_kv_cell_ext during state read (#20273)

sprayandwipe committed 5 days ago

Verified 6b10a82c

model : wire up Qwen3.5/Qwen3.5MoE tensors for NVFP4 support (#20506)

michaelw9999 committed 6 days ago

Verified d23355af

metal : add FA specialization for HSK = 320, HSV = 256 (#20549)

ggerganov committed 6 days ago

Verified b30a5fdf

ci : move self-hosted workflows to separate files (#20540)

ggerganov committed 6 days ago

Verified b4768955

docker : force Python 3.13 in Vulkan container (#20530)

gguillemas committed 6 days ago

Verified fc350fdf

ci : try to optimize some jobs (#20521)

netrunnereve committed 6 days ago

Verified 3a6f0599

hexagon: Q4_0 and MXFP4 repack fixes (#20527)

max-krasnyansky committed 6 days ago

Verified 609ea500

ci : reduce webgpu tests timeout to 900s (#20538)

ggerganov committed 6 days ago

Verified 9f774e45

mtmd: add llama-mtmd-debug binary (#20508)

ngxson committed 6 days ago

Verified 94d02622

add op gated_delta_net (#20455)

arthw committed 6 days ago

Verified a93c0ef0

webui: restore code preview iframe origin isolation (#20477)

Chedrian07 committed 6 days ago

Verified 710878a7

scripts : remove get-wikitext-103.sh (#20543)

angt committed 6 days ago

Verified 0685848b

scripts : update get-hellaswag.sh and get-winogrande.sh (#20542)

angt committed 6 days ago

Verified 0024a69b

ggml : add native AVX512-FP16 support for F16 operations (#20529)

angt committed 6 days ago

Verified d0b79aaa

Use fp32 in cuBLAS V100 to avoid overflows, env variables to override cuBLAS compute type (#19959)

wallentri88 committed 6 days ago

Verified f2c0dfb7

ggml : add OpenVINO backend (#15307)

wine99 committed 6 days ago

Verified 9789c4ec

vendor : update cpp-httplib to 0.37.2 (#20484)

angt committed 6 days ago

Verified 77e20cc1

Fix data race in CUDA's "cpy" kernel (influences GGML's DUP, CONT operations). (#20507)

Exile333 committed 6 days ago

Verified 5a32a9b8

opencl: fix l2_norm (#20480)

lhez committed 6 days ago

Verified 3b439504

tools : enable kvu in perplexity for hellaswag, winogrande, multiple-choice (#19954)

angt committed 7 days ago

Verified 463b6a96

graph : remove redundant GDN state transposes (#20443)

ggerganov committed 7 days ago

Verified e30f1fdf

common/parser: gracefully handle undetected tool parser, print error message. (#20286)

pwilkin committed 7 days ago

Verified 1430c359

llama : fix pooling assertion crash in chunked GDN detection path (#20468)

ZeroV0LT committed 7 days ago

Verified f17b3be6

server: reset counter related to kill-switch on client error (#20513)

SoftwareRenderer committed 7 days ago

Verified d7ba99c4

Older