ggerganov/llama.cpp

Pull Requests Commits

metal : remove the in-kernel dequantization from the FA kernels

ggerganov committed 8 hours ago

06337b61

test : make the FA V-is-view-of-K case a test case parameter

ggerganov committed 9 hours ago

9126f3a8

metal : rename the FA dequant_f16 identifiers to kv_f16

ggerganov committed 10 hours ago

101428bb

pi : avoid backend mentions in test-backend-ops comments

ggerganov committed 10 hours ago

275768fb

test : remove backend-specific wording from test-backend-ops comments

ggerganov committed 10 hours ago

34ad8da7

metal : skip the redundant V dequant when V is a view of K

ggerganov committed 10 hours ago

cd644cf1

metal : dequantize q4_0, q4_1, q5_0 and q5_1 KV to f16 before flash attention

ggerganov committed 10 hours ago

74c13eea

metal : launch the FA KV dequant kernel separately for K and V

ggerganov committed 10 hours ago

ad615b62

metal: dequantize q8_0 KV to f16 before flash attention

ggerganov committed 10 hours ago

5d6b00b7

metal : dequantize q8_0 using packed types (#27370)

ggerganov committed 12 hours ago

Verified 8ef78e64

server: add dedup-cache-models preset option (#27346)

ngxson committed 15 hours ago

Verified ee4c505a

vulkan: tiled transpose for 0<->2 permuted CONT (#26585)

kh0pper committed 15 hours ago

Verified 98d1e92c

ggml-webgpu: add mulmat with overlapping src0/src1 (e.g., for minimax-01) (#27321)

yomaytk committed 16 hours ago

Verified 5112b973

ci : add attestation for signed release artifacts (#25933)

quartzjer committed 16 hours ago

Verified 0adcc3bb

tests: skip test-unicode build on win32/BUILD_SHARED_LIBS (#27336)

jeffbolznv committed 17 hours ago

Verified 79fe7991

gguf-py : add size guards to GGUFReader (#27188)

oakkaya committed 17 hours ago

Verified 0329fcda

server: (cosmetic) do not print cmd_child_to_router messages [no release] (#27347)

ngxson committed 1 day ago

Verified 6d054983

opencl: fix WAR race in the generic FA tile kernels when the WG spans subgroups (#26434)

wanghqc committed 1 day ago

Verified 60addddf

mtmd: deepseek-ocr SAM ggml_conv_2d with the im2col kept in F32 (#26727)

sfallah committed 1 day ago

Verified 9731ad3f

RPC: populate use_count to enable fusion inside backends (#27142)

am17an committed 1 day ago

Verified af517262

ui: Services consolidation refactor (#27239)

allozaur committed 1 day ago

Verified 3dc7285b

ui: Stores consolidation refactor (#27238)

allozaur committed 1 day ago

Verified fdf4c646

common: share thread pools when `n_threads` differ (#27138)

nikwen committed 1 day ago

Verified 04b56914

quant : Optimise memory usage by evicting weights after processing each layer (#22877)

EAddario committed 1 day ago

Verified 05967042

sycl: honor GGML_HINT_SRC0_IS_HADAMARD (#27298)

Titaniumtown committed 1 day ago

Verified 0882c7bc

unicode : include '~' in collapsed symbol class (#26972)

tarruda committed 1 day ago

Verified afd439df

readme : update status badges + regen AUTHORS (#27317)

ggerganov committed 1 day ago

Verified 169e4a7f

ci : Update OpenVINO to 2026.3, skip nemotron-h rollback test (#27292)

wine99 committed 1 day ago

Verified 9d77fa17

ci: add Windows ARM64 CUDA support to the manual workflow (#27300)

shivamkumard-ctrl committed 1 day ago

Verified c0296022

mtmd: fix LFM2 image tiling threshold (#27057)

BlackFoil committed 1 day ago

Verified 7acdbb1f

Older