ngxson/llama.cpp

Pull Requests Commits

hex-dma: introduce direct-mapped DMA cache that is better suited for HVX FA mask handling (#29282)

max-krasnyansky committed 4 hours ago

Verified e6ab7c1a

HIP : optimize IQ2/IQ3 (`__vsub4` `__vcmpne4`) using SWAR (#27962)

yanjs committed 6 hours ago

Verified f46bc30c

jinja : fix dangling reference warning in for_statement (#29279)

ggerganov committed 7 hours ago

Verified 709fe755

opencl: add bin kernel `kernel_gemm_noshuffle_q4_k_q8_1_dp4a_ila_a8_bin` (#29056)

shaofeiqi committed 7 hours ago

Verified d5f66492

server: fix router eviction races with the existing queue (#29217)

ServeurpersoCom committed 7 hours ago

Verified 99199111

server: do not pass log file to children (#29212)

dfriehs committed 7 hours ago

Verified bbf99b1b

server: support input_image in function_call_output (#20663) (#22575)

Empressia committed 7 hours ago

Verified 4098fdc9

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

fish-jiang committed 10 hours ago

Verified 4ceb1719

mtmd: add various sanity checks (#29276)

ngxson committed 10 hours ago

Verified 73c941b1

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

mdegans committed 11 hours ago

Verified 0f8a414b

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

bartowski1182 committed 12 hours ago

Verified f95b0d95

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

frobnitzem committed 13 hours ago

Verified c350a40b

ui : Accept WEBM video files (#28622)

EpicEric committed 13 hours ago

Verified 9b421fa9

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

ngxson committed 13 hours ago

Verified 348f853b

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

erusev committed 13 hours ago

Verified 217f81c2

spec : support DFlash for HunyuanOCR (#28890)

wendadawen committed 13 hours ago

Verified 828fdf28

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

AesSedai committed 14 hours ago

Verified bfd73a87

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

miyanyan committed 14 hours ago

Verified a60f9aea

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

NickM-27 committed 19 hours ago

Verified 7ab4ee7b

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

ykhrustalev committed 19 hours ago

Verified 0ee9435b

Add close button to UI toasts (#28246)

agustinmista committed 19 hours ago

Verified 8cfc315a

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

shaofeiqi committed 20 hours 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 22 hours ago

Verified c550d2f6

hexagon: new HMX-optimized GATED_DELTA_NET (#29199)

max-krasnyansky committed 1 day ago

Verified 58367713

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

angt committed 1 day ago

Verified ff0dbb97

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

masterFoad committed 1 day ago

Verified fb34fc26

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

ggerganov committed 1 day ago

Verified c641dfa8

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

ggerganov committed 1 day ago

Verified 96550613

cuda: fix sm_70 tile compilation error (#29224)

lingyezhixing committed 1 day ago

Verified b1c2863e

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

pwilkin committed 1 day ago

Verified f4e276a2

Older