ngxson/llama.cpp

Pull Requests Commits

server: allow preset to set log file

ngxson committed 47 days ago

8e87cb98

metal: add the missing f32 x bf16 mul_mv variants (#28741)

ServeurpersoCom committed 47 days ago

Verified 95753896

CUDA: enable sparse-fa for dsv4 prefill (again) (#29298)

am17an committed 47 days ago

Verified dc9879cf

server: fix token counting API crash on sleep (#29309)

willweimike committed 47 days ago

Verified 42916d83

jinja : parse unary +/- before variables (#29244)

cs-fisha committed 48 days ago

Verified 4e416ee7

metal : key the fa-vec tuned table by family instead of SKU (#29075)

forforever73 committed 48 days ago

Verified ee3ecce0

server: accept OpenAI video_url content type and data: video URIs (#27921)

calebrio02 committed 48 days ago

Verified 057494f9

server: Dedup the draft HF model via dedup-cache-models (#27934)

DreamingWater committed 48 days ago

Verified bcbc936a

ci : fix build-cmake runner target (#29299)

CISC committed 48 days ago

Verified 26758d38

model-conversion : add causal-compare-logits recipe (#29305)

danbev committed 48 days ago

Verified 18f9f7be

model : support Gemma4 DSpark draft backbone (#29226)

hthadicherla committed 48 days ago

Verified 633733d0

ci : run python (jinja) test (#29302)

CISC committed 48 days ago

Verified 86b2daa7

make-release : update summary prompt

ggerganov committed 48 days ago

183d2a04

ggerganov committed 48 days ago

45062d40

ggml : bump version to 0.25.0 (ggml/1635)

ggerganov committed 48 days ago

503549c5

sycl : fix compile warnings

ggerganov committed 48 days ago

e97545d9

vulkan: add IQ4_XS MMQ/MMV matmul kernels (#28415)

pwilkin committed 48 days ago

Verified b1ff4ca2

ggml-meta: resolve multi buffer views (#29266)

0cc4m committed 48 days ago

Verified 94256114

cuda: top-k MoE should always fire (#28432)

am17an committed 48 days ago

Verified 1a679828

sycl : support new UT case for mul_mat_hadamard fp16 (#29218)

arthw committed 48 days ago

Verified 384a534c

sycl: extend MMVQ GLU fusion, add rms_norm+scale and ssm_conv+silu fusions (#28931)

anantshri committed 48 days ago

Verified 5e48b310

sycl : support op get_rows_back, only support fp32/fp16 (#25266)

arthw committed 48 days ago

Verified 4d7d7703

vulkan: hide internal symbols to prevent duplicate-dlopen state destruction (#29139)

ewintr committed 48 days ago

Verified 08b1d2ae

sampler: reduce the size of the probe (#29285)

max-krasnyansky committed 48 days ago

Verified 441df11f

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

max-krasnyansky committed 48 days ago

Verified e6ab7c1a

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

yanjs committed 48 days ago

Verified f46bc30c

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

ggerganov committed 48 days ago

Verified 709fe755

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

shaofeiqi committed 48 days ago

Verified d5f66492

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

ServeurpersoCom committed 48 days ago

Verified 99199111

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

dfriehs committed 48 days ago

Verified bbf99b1b

Older