ggml-org/llama.cpp

Pull Requests Commits
unicode : MSVC regex fix (#19340)
Iemand005 committed 46 days ago
Verified 06bf3796
Kimi-Linear support (backend agnostic + MLA KV cache) (#18755)
ymcki committed 46 days ago
Verified 3688c4f5
vulkan: For coopmat2 FA, use fp16 accumulators for the final result (#19376)
jeffbolznv committed 46 days ago
Verified 1946e46f
vulkan: make FA mask/softcap enables spec constants (#19309)
jeffbolznv committed 46 days ago
Verified f9bd518a
metal : skip loading all-zero mask (#19337)
ggerganov committed 46 days ago
Verified 7fcf1ef4
llama : rename llama-sampling to llama-sampler (#19363)
danbev committed 46 days ago
Verified e696cfc0
cuda : cuda graphs now compare all node params (#19383)
ggerganov committed 46 days ago
Verified 3e216476
metal : adaptive CPU/GPU interleave based on number of nodes (#19369)
ggerganov committed 46 days ago
Verified 22cae832
vulkan: Preprocess FA mask to detect all-neg-inf and all-zero. (#19281)
jeffbolznv committed 47 days ago
Verified 449ec2ab
benches : update models + numbers (#19359)
ggerganov committed 47 days ago
Verified 3795cc1e
docker : fix vulkan build (#19352)
CISC committed 47 days ago
Verified b828e18c
vendor : update BoringSSL to 0.20260204.0 (#19333)
angt committed 47 days ago
Verified a4ea7a18
metal : add diag (#19330)
ggerganov committed 47 days ago
Verified 7a4f97d1
vulkan: fix GPU deduplication logic. (#19222)
okuvshynov committed 47 days ago
Verified a498c75a
vulkan: Set k_load_shmem to false when K is too large (#19301)
jeffbolznv committed 47 days ago
Verified 3409ab84
vulkan: fix non-contig rope (#19299)
jeffbolznv committed 47 days ago
Verified c342c3b9
metal : add missing includes (#19348)
will-lms committed 47 days ago
Verified af252d07
vendor : add missing llama_add_compile_flags (#19322)
CISC committed 47 days ago
Verified 11fb327b
vendor: update cpp-httplib version (#19313)
taronaeo committed 47 days ago
Verified e6e934c5
codeowners : add danbev for examples/debug (#19332)
danbev committed 47 days ago
Verified b536eb02
debug: make common_debug_print_tensor readable (#19331)
ngxson committed 47 days ago
Verified e0c93af2
ci : fix sanitize workflow to enable ggml sanitizers too (#19323)
ggerganov committed 48 days ago
Verified 423bee46
model: (qwen3next) correct vectorized key_gdiff calculation (#19324)
ngxson committed 48 days ago
Verified 8abcc70a
tests : add non-cont, inplace rope tests (#19296)
ggerganov committed 48 days ago
Verified eaba92c3
model-conversion : add tensor-info.py utility (#18954)
danbev committed 48 days ago
Verified 6ab881b7
spec : fix the check-rate logic of ngram-simple (#19261)
ggerganov committed 48 days ago
Verified d838c22b
completion : simplify batch (embd) processing (#19286)
danbev committed 48 days ago
Verified 25f40ca6
ggml-virtgpu: make the code thread safe (#19204)
kpouget committed 48 days ago
Verified 015deb90
ggml-cpu: use LUT for converting e8->f32 scales on x86 (#19288)
am17an committed 48 days ago
Verified 2ceda3f6
metal : add solve_tri (#19302)
ggerganov committed 48 days ago
Verified 44008ce8
Terms Of Use Privacy Policy Refund Policy Impressum