ggerganov/llama.cpp

Pull Requests Commits

fix: rpc-server cache may not work in Windows environments (#22394)

unraido committed 19 days ago

Verified ceaf47c4

Fix type casting for unaccounted memory calculation (#22424)

rankaiyx committed 19 days ago

Verified 42401c72

download : prefer q8_0 when q4_k not available (#22428)

ggerganov committed 19 days ago

Verified e940b3d4

model : remove duplicate wo_s scale after build_attn (Qwen3, LLaMA) (#22421)

ynankani committed 19 days ago

Verified 0f1bb602

convert : remove input_scale for dequantized fp8 modelopt (#22356)

CISC committed 19 days ago

Verified d13540be

ggml : use 64 bytes aligned tile buffers (#21058)

angt committed 19 days ago

Verified f84270ea

common: fix missing exports in llama-common (#22340)

max-krasnyansky committed 19 days ago

Verified 5594d132

pr2wt : symlink .pi (#22386)

ggerganov committed 20 days ago

Verified f5357743

add performance-portable tuning for register-tile and subgroup matmul (#22241)

SharmaRithik committed 20 days ago

Verified 06a811d0

Fix recurrent state serialization for partial reads and writes (#22362)

gaugarg-nv committed 20 days ago

Verified 78433f60

Github: set meta backend code owner (#22388)

JohannesGaessler committed 20 days ago

Verified 7ec36aa8

CUDA: better coalesce data-access for contiguous concat (#22330)

ORippler committed 20 days ago

Verified b1a5bd4e

ggml-cpu : re-enable fast gelu_quick_f16 (#22339)

CISC committed 20 days ago

Verified 0c6ee1ca

ggml-cpu: optimize avx2 q6_k (#22345)

netrunnereve committed 20 days ago

Verified 2dd84169

opencl: add iq4_nl support (#22272)

lhez committed 20 days ago

Verified f454bd7e

hexagon: guard HMX clock request for v75+ platforms (#22377)

trivikram-reddy1 committed 21 days ago

Verified b760272f

chat: fix handling of space in reasoning markers (#22353)

pwilkin committed 21 days ago

Verified dcad77cc

spec : fix vocab compat checks (#22358)

ggerganov committed 21 days ago

Verified 98dc1418

CUDA: reduce MMQ stream-k overhead (#22298)

JohannesGaessler committed 21 days ago

Verified 9725a313

metal : optimize Metal Tensor API usage for GGML_OP_MUL_MAT (#20962)

Developer-Ecosystem-Engineering committed 21 days ago

Verified d1649047

llama-quant : default ftype param `Q5_1` --> `Q8_0` (#20828)

ddh0 committed 21 days ago

Verified 9d34231b

gitignore : add .pi + personal SYSTEM.md (#22316)

ggerganov committed 21 days ago

Verified 8ea8fee9

[SYCL] Optimize Q4_0 mul_mat for Arc770, add scripts (#22291)

arthw committed 21 days ago

Verified eddd7a13

ggml-webgpu: support for SSM_SCAN and disable set_rows error checking (#22327)

reeselevine committed 21 days ago

Verified dd2914dc

parser: fix structured output bug (#22302)

pwilkin committed 22 days ago

Verified 0adede86

Hexagon: Bump HMX Frequency to Max Corner (#22334)

trivikram-reddy1 committed 22 days ago

Verified 361fe72a

CI Snapdragon: Switch ubuntu-latest to ubuntu-slim runner (#22303)

shreyajn committed 22 days ago

Verified a702f395

ggml-webgpu: enable FLASH_ATTN_EXT on browser without subgroup matrix (#22199)

ArberSephirotheca committed 22 days ago

Verified 13d36cf8

hexagon: use DIRID 13 in libggml-htp.inf for modern InfVerif (#22306)

mengshengwu committed 22 days ago

Verified f65bc34c

metal : print GPU description (#22318)

ggerganov committed 22 days ago

Verified 15fa3c49

Newer Older