ggerganov/llama.cpp

Pull Requests Commits

profiler: output all tensor names

max-krasnyansky committed 19 days ago

00332635

profiler: initial support for profiling graph ops

max-krasnyansky committed 19 days ago

b3d6ed39

mtmd : support home-cooked Mistral Small Omni (#14928)

ngxson committed 19 days ago

Verified 1bb4f433

fix: added a normalization step for MathJax-style \[\] and \(\) delimiters (#16599)

ServeurpersoCom committed 19 days ago

Verified 683fa6ba

sycl : add ARANGE operator (#16362)

GittyBurstein committed 19 days ago

Verified b22572e9

CANN: format code using .clang-format (#15863)

noemotiovon committed 19 days ago

Verified 7a50cf38

common : Update the docs on -t --threads (#16236)

takasurazeem committed 19 days ago

Verified 6f5d9246

ggml-cpu: replace putenv with setenv for const-correctness (#16573)

otegami committed 19 days ago

Verified adc9b60f

SYCL: Add GGML_OP_MEAN operator support (#16009)

yael-works committed 19 days ago

Verified ee50ee1e

gguf-py : add support for endian conversion of BF16 data (#16594)

AlekseiNikiforovIBM committed 19 days ago

Verified 7adc79c0

cpu : add FLOOR, CEIL, ROUND and TRUNC unary operators (#16083)

safranowith committed 19 days ago

Verified 466c1911

opencl: add q8_0 mm support (#16469)

lhez committed 20 days ago

Verified 0cb7a068

opencl: fix FA for f32 (#16584)

lhez committed 20 days ago

Verified d93f8439

Add server-driven parameter defaults and syncing (#16515)

allozaur committed 20 days ago

Verified f9fb33f2

metal: optimise `GGML_OP_SUM` (#16559)

cern1710 committed 20 days ago

Verified f4ce81c4

server : fix img token logs (#16595)

ggerganov committed 20 days ago

Verified 17304cbc

llama-quant: add support for mmproj (#16592)

ngxson committed 20 days ago

Verified 3e3cb19f

CUDA: Changing the CUDA scheduling strategy to spin (#16585)

JTischbein committed 20 days ago

Verified 5acd4554

server : fix mtmd checkpoints (#16591)

ggerganov committed 20 days ago

Verified 554fd578

metal : avoid using Metal's gpuAddress property (#16576)

ggerganov committed 21 days ago

Verified fa882fd2

vulkan: Add ACC_TYPE_VEC2 implementation (#16203)

SavicStefan committed 21 days ago

Verified ffa05903

CUDA + openCL: fix bug in accessing rms_norm->src while doing fusion (#16577)

am17an committed 21 days ago

Verified 120bf704

vulkan: Support FA with K/V in F32 (#16543)

jeffbolznv committed 21 days ago

Verified 4258e0cf

vulkan: Improve build time for MSVC (#16545)

jeffbolznv committed 21 days ago

Verified 7ea15bb6

CUDA: enable FA for FP32 KV cache (#16546)

JohannesGaessler committed 21 days ago

Verified 9c7185dd

CUDA: use fastdiv + ggml_cuda_mad for mmvf (#16557)

am17an committed 21 days ago

Verified 1ee9d0b4

CUDA: add fp kernel for larger batch size MoE (#16512)

am17an committed 21 days ago

Verified 48e2fa9f

cuda : remove legacy copy-op pointer indirection code (#16485)

anavp-nvidia committed 21 days ago

Verified 5b6913c4

server : dynamic token limit for prompt cache (#16560)

ggerganov committed 21 days ago

Verified bc07349a

metal : FA support F32 K and V and head size = 32 (#16531)

ggerganov committed 21 days ago

Verified e60f241e

Older