ggml-org/llama.cpp

Pull Requests Commits

jinja : fix dangling reference warning in for_statement

ggerganov committed 26 days ago

f0fffd6c

vulkan: add Intel Xe flash attention optimization kernels (2/3, Xe-LPG Plus/Xe2/Xe3) (#24406)

fish-jiang committed 27 days ago

Verified 4ceb1719

mtmd: add various sanity checks (#29276)

ngxson committed 27 days ago

Verified 73c941b1

metal : gate mul_mm_id src1 rescale behind ggml_prec (#29029)

mdegans committed 27 days ago

Verified 0f8a414b

ggml : IQ1_M build prefix sums once per block (#28706)

bartowski1182 committed 27 days ago

Verified f95b0d95

Performance tune for gemma4-26b-a4b flash attention shape. (#28450)

frobnitzem committed 27 days ago

Verified c350a40b

ui : Accept WEBM video files (#28622)

EpicEric committed 27 days ago

Verified 9b421fa9

jinja: use const for statement::execute and ::visit (#29271)

ngxson committed 27 days ago

Verified 348f853b

server: Add support for binding to multiple addresses (#28690)

erusev committed 27 days ago

Verified 217f81c2

spec : support DFlash for HunyuanOCR (#28890)

wendadawen committed 27 days ago

Verified 828fdf28

convert: add MiMo-V2.6 support (#29257)

AesSedai committed 27 days ago

Verified bfd73a87

cmake : allow repeated find_package calls for llama (#29228)

miyanyan committed 27 days ago

Verified a60f9aea

chat : Fix Muse Glimmer tool-call first parser error (#29242)

NickM-27 committed 27 days ago

Verified 7ab4ee7b

ci : publish snapdragon builds in release workflow (#29007)

ykhrustalev committed 27 days ago

Verified 0ee9435b

Add close button to UI toasts (#28246)

agustinmista committed 27 days ago

Verified 8cfc315a

opencl: add A8 Q4_0 non-MoE dp4a binary kernel (#29055)

shaofeiqi committed 27 days ago

Verified ec5a12b8

ci : update Level Zero SDK to v1.33.1 and enable the L0/oneDNN CMake flags in the SYCL job (#29230)

Asahi-Prv committed 27 days ago

Verified c550d2f6

hexagon: new HMX-optimized GATED_DELTA_NET (#29199)

max-krasnyansky committed 27 days ago

Verified 58367713

vendor : update cpp-httplib to 0.57.1 (#29239)

angt committed 27 days ago

Verified ff0dbb97

metal : fix mask bounds in flash attention block pre-pass (#29220)

masterFoad committed 27 days ago

Verified fb34fc26

test-save-load-state : compare logits with NMSE and feed expected tokens (#29238)

ggerganov committed 27 days ago

Verified c641dfa8

llama-context : report graph inputs and input tensors during sched reserve (#26625)

ggerganov committed 28 days ago

Verified 96550613

cuda: fix sm_70 tile compilation error (#29224)

lingyezhixing committed 28 days ago

Verified b1c2863e

ggml-cuda : convert contiguous tensors four elements at a time (#29155)

pwilkin committed 28 days ago

Verified f4e276a2

cuda : accelerate conv2d with implicit GEMM (#29135)

leejet committed 28 days ago

Verified e6cef815

ggml : fix dimension and stride truncation in ggml_permute (#29227)

leejet committed 28 days ago

Verified c21284cd

vendor : update cpp-httplib to 0.57.0 (#29214)

angt committed 28 days ago

Verified 6f41ac59

docker : bump cuda to 13.4.1 (#29207)

CISC committed 28 days ago

Verified ec91ab5a

sycl : support gated DSV4_HC_PRE and optional HC_POST comb matrix (#29132)

cwriter committed 28 days ago

Verified bb3c853c

sycl : pinned memory use right device context instead of 0 (#28895)

lslusarczyk committed 28 days ago

Verified af911149

Older