ngxson/llama.cpp

Pull Requests Commits

ci : fix sanitize workflow to enable ggml sanitizers too (#19323)

ggerganov committed 16 days ago

Verified 423bee46

model: (qwen3next) correct vectorized key_gdiff calculation (#19324)

ngxson committed 16 days ago

Verified 8abcc70a

tests : add non-cont, inplace rope tests (#19296)

ggerganov committed 16 days ago

Verified eaba92c3

model-conversion : add tensor-info.py utility (#18954)

danbev committed 16 days ago

Verified 6ab881b7

spec : fix the check-rate logic of ngram-simple (#19261)

ggerganov committed 16 days ago

Verified d838c22b

completion : simplify batch (embd) processing (#19286)

danbev committed 16 days ago

Verified 25f40ca6

ggml-virtgpu: make the code thread safe (#19204)

kpouget committed 16 days ago

Verified 015deb90

ggml-cpu: use LUT for converting e8->f32 scales on x86 (#19288)

am17an committed 16 days ago

Verified 2ceda3f6

metal : add solve_tri (#19302)

ggerganov committed 17 days ago

Verified 44008ce8

ci : add sanitizer runs for server (#19291)

ggerganov committed 17 days ago

Verified 6a9bf2f7

sampling : delegate input allocation to the scheduler (#19266)

ggerganov committed 17 days ago

Verified faa1bc26

vulkan: disable coopmat1 fa on Nvidia Turing (#19290)

0cc4m committed 17 days ago

Verified 32b17abd

CUDA: use mmvq for mul-mat-id for small batch sizes (#18958)

am17an committed 17 days ago

Verified 8bece2eb

models : remove unnecessary cont in openelm (#19289)

CISC committed 17 days ago

Verified a6fd8ca1

metal : minor cleanup (#19251)

ggerganov committed 17 days ago

Verified c55bce41

CUDA: Fix loop unrolling for BW in mul_mat_q_stream_k_fixup (#19053)

ORippler committed 17 days ago

Verified 1f1e57f2

ggml: added cleanups in ggml_quantize_free (#19278)

noctrex committed 17 days ago

Verified e9a859db

cuda : revert CUDA_SCALE_LAUNCH_QUEUES override until investigated (#19227)

gaugarg-nv committed 17 days ago

Verified 41e3f026

vocab: add Falcon-H1-Tiny-Coder FIM tokens (#19249)

vhsw committed 17 days ago

Verified 1efb5f7a

spec : simplify time measurement using common_time_meas (#19262)

ggerganov committed 17 days ago

Verified aeb827a3

opencl: refactor some ops, concat, repeat, tanh and scale (#19226)

lhez committed 17 days ago

Verified 91ea44e8

jinja : add missing 'in' test to template engine (#19004) (#19239)

sidmohan0 committed 18 days ago

Verified 0dfcd3b6

mtmd: add min/max pixels gguf metadata (#19273)

ngxson committed 18 days ago

Verified 07a7412a

ggml-cpu: FA split across kv for faster TG (#19209)

am17an committed 18 days ago

Verified 9f682fb6

server: print actual model name in 'model not found" error (#19117)

teto committed 18 days ago

Verified a3fa0358

ci: add test-backend-ops test for CPU (#19268)

am17an committed 18 days ago

Verified 15818ac4

Remove support for Nvidia & AMD GPU, because the oneAPI plugin for Nvidia & AMD GPU is unavailable: download/installation channels are out of work. (#19246)

arthw committed 18 days ago

Verified bf38346d

sycl: implement GGML_OP_TOP_K (#19242)

tdevelope committed 18 days ago

Verified 4d5e9726

metal : support virtual devices (#18919)

ggerganov committed 18 days ago

Verified 6fdddb49

model-conversion : add debug option to conversion script (#19265)

danbev committed 18 days ago

Verified 6156ae51

Newer Older