ggerganov/llama.cpp

Pull Requests Commits

Merge remote-tracking branch 'origin/master' into xsn/dots3-note

ngxson committed 51 days ago

a3f308aa

metal : clamp K extent in tensor API mat-mat kernel for K not a multiple of 32 (#27450)

ggerganov committed 51 days ago

Verified a3027337

opencl: fix q6_K flat mul_mat for Adreno A6x/A7x GPUs with older E031 compilers (#26476)

wanghqc committed 51 days ago

Verified 6503355d

opencl: fix local size for norm (#27339)

lhez committed 51 days ago

Verified 6b4fa88a

ui: Stores split refactor (#27240)

allozaur committed 51 days ago

Verified 521a64cd

mtmd: add --mmproj-device argument (#23255)

Interpause committed 51 days ago

Verified 681c29d3

model : support DSpark for LFM2 models (#27383)

tdakhran committed 52 days ago

Verified 07822bdd

vulkan: FA MMQ should use fp32 for Q quantization calculations (#27413)

jeffbolznv committed 52 days ago

Verified 78ec4c37

metal : dequant kv cache only for large batches (#27438)

ggerganov committed 52 days ago

Verified 63b64a50

CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (#26678)

ORippler committed 52 days ago

Verified bf0040e1

server: (router) lazy-load startup_models after main setup (#27424)

ngxson committed 52 days ago

Verified 9855ad69

server : fix --docker-repo being treated as router mode (#27416)

AriBandyo committed 52 days ago

Verified 8a832e4b

CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover (#26079)

praneshgo committed 52 days ago

Verified 2b562109

common : gracefully fallback on unsupported regex patterns in JSON schema (#26939)

aldehir committed 52 days ago

Verified dc64a162

metal : dequantize quantized KV to F16 before flash attention (#27390)

ggerganov committed 52 days ago

Verified 70aff252

Revert "tensor-split meta backend fixes (#26502)" (#27433)

ggerganov committed 52 days ago

Verified f20395da

ggml: fix backend split scheduler race condition (#26040)

0cc4m committed 52 days ago

Verified 84979813

convert: fix get block count error for Nemotron 3 Ultra (#27101)

rockchenn committed 52 days ago

Verified a3b1effc

ggml-cuda: provide static workspace for cuBLAS handles (#26574)

heislera763 committed 52 days ago

Verified d9b6be07

graph : create V as a view of K in the k_iswa build_attn (#27392)

ggerganov committed 52 days ago

Verified 929d47a3

spec : avoid binding reference to null pointer (#27404)

ggerganov committed 52 days ago

Verified f466cfa3

vulkan : add source groups for shaders (#26666)

mtavenrath committed 52 days ago

Verified 2cfdb5fc

opencl: make the MoE expert scatter deterministic (#26464)

wanghqc committed 52 days ago

Verified 9ee9fc04

tensor-split meta backend fixes (#26502)

max-krasnyansky committed 52 days ago

Verified d59d455f

hexagon: fix FA HMX queue ordering and pack the rescale D matrices (#27042)

njsyw1997 committed 52 days ago

Verified 990e3bfe

opencl: port fused ssm_scan kernel (Mamba-2, d_state in {128, 256}) to GPU (#26439)

wanghqc committed 52 days ago

Verified b062ba73

ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (#26860)

ServeurpersoCom committed 52 days ago

Verified cd644c39

server: refactor sleep handling, allow access /metrics during sleep (#27376)

ngxson committed 52 days ago

Verified 947fd9bb

move to a new llama_kv_cache_dsa_iswa

ngxson committed 52 days ago

d9de2b20

server : make models endpoints private when authentication is enabled (#26347)

s0mecode committed 52 days ago

Verified ee0ea03a

Older