ggml-org/llama.cpp

Pull Requests Commits

models : optimizing qwen3next graph

ggerganov committed 2 days ago

Verified 6d959971

metal : adaptive CPU/GPU interleave based on number of nodes (#19369)

ggerganov committed 2 days ago

Verified 22cae832

vulkan: Preprocess FA mask to detect all-neg-inf and all-zero. (#19281)

jeffbolznv committed 2 days ago

Verified 449ec2ab

benches : update models + numbers (#19359)

ggerganov committed 2 days ago

Verified 3795cc1e

docker : fix vulkan build (#19352)

CISC committed 2 days ago

Verified b828e18c

vendor : update BoringSSL to 0.20260204.0 (#19333)

angt committed 2 days ago

Verified a4ea7a18

metal : add diag (#19330)

ggerganov committed 2 days ago

Verified 7a4f97d1

vulkan: fix GPU deduplication logic. (#19222)

okuvshynov committed 2 days ago

Verified a498c75a

vulkan: Set k_load_shmem to false when K is too large (#19301)

jeffbolznv committed 2 days ago

Verified 3409ab84

vulkan: fix non-contig rope (#19299)

jeffbolznv committed 2 days ago

Verified c342c3b9

metal : add missing includes (#19348)

will-lms committed 2 days ago

Verified af252d07

vendor : add missing llama_add_compile_flags (#19322)

CISC committed 3 days ago

Verified 11fb327b

vendor: update cpp-httplib version (#19313)

taronaeo committed 3 days ago

Verified e6e934c5

codeowners : add danbev for examples/debug (#19332)

danbev committed 3 days ago

Verified b536eb02

debug: make common_debug_print_tensor readable (#19331)

ngxson committed 3 days ago

Verified e0c93af2

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

ggerganov committed 3 days ago

Verified 423bee46

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

ngxson committed 3 days ago

Verified 8abcc70a

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

ggerganov committed 3 days ago

Verified eaba92c3

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

danbev committed 3 days ago

Verified 6ab881b7

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

ggerganov committed 3 days ago

Verified d838c22b

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

danbev committed 3 days ago

Verified 25f40ca6

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

kpouget committed 4 days ago

Verified 015deb90

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

am17an committed 4 days ago

Verified 2ceda3f6

metal : add solve_tri (#19302)

ggerganov committed 4 days ago

Verified 44008ce8

ci : add sanitizer runs for server (#19291)

ggerganov committed 4 days ago

Verified 6a9bf2f7

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

ggerganov committed 4 days ago

Verified faa1bc26

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

0cc4m committed 4 days ago

Verified 32b17abd

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

am17an committed 4 days ago

Verified 8bece2eb

models : remove unnecessary cont in openelm (#19289)

CISC committed 4 days ago

Verified a6fd8ca1

metal : minor cleanup (#19251)

ggerganov committed 4 days ago

Verified c55bce41

Older