ggerganov/llama.cpp

Pull Requests Commits

ngxson committed 3 hours ago

a5a49f9e

Merge branch 'master' into xsn/chat_input_parts

ngxson committed 4 hours ago

8fcbc494

ggml : simplify backend search path error handling (#28367)

angt committed 8 hours ago

Verified 1b70350c

docs : update zDNN supported ops (#30472)

taronaeo committed 8 hours ago

Verified d71e324a

kleidiai: add SME2 Q4_K and Q6_K matmul support (#29219)

chaxu01 committed 10 hours ago

Verified 45451c36

vulkan: Optimize pool_2d when the pooling window is large (#30119)

jeffbolznv committed 11 hours ago

Verified 77c4540c

meta: recreate all subgraphs when the arena is reallocated (#30453)

Xanewok committed 11 hours ago

Verified ecdd29e8

hexagon: add configurable HTP DCVS/HMX power mode (#30423)

pdhinaka committed 11 hours ago

Verified ec0bfac1

graph: pad the KQ bias of the no-cache attention (#30446)

ServeurpersoCom committed 12 hours ago

Verified ace20f67

llama : do not map foreign ranges into a backend buffer (#30405)

ggerganov committed 12 hours ago

Verified b9f3599f

model: load YaRN beta metadata for laguna models (#30331)

accaldwell committed 12 hours ago

Verified f441d2d9

graph: skip the LM head for pooled embeddings (#30445)

ServeurpersoCom committed 12 hours ago

Verified cb55ffe6

tests : consolidate context state tests into test-llama-context (#30023)

ggerganov committed 13 hours ago

Verified a17e94d2

llama : fix log message (#30452)

ggerganov committed 13 hours ago

Verified b476d9a7

vulkan: allow multiple devices to access gpu-visible host memory (#29741)

0cc4m committed 13 hours ago

Verified e0bcaccb

ggml-cpu : add AVX2 dot product for Q2_0 (#30447)

homeofe committed 14 hours ago

Verified 0a731cce

cuda : add GGML_CUDA_GRAPH_DEBUG and read env vars once (#30327)

ggerganov committed 14 hours ago

Verified f851e244

cuda : fix compiler warnings in mmq-load-tiles.cuh (#30443)

angt committed 14 hours ago

Verified bfe65e1d

vulkan : split matmul dispatch when it would exceed maxComputeWorkGroupCount (#29533)

losewayy committed 15 hours ago

Verified 411d5485

spec : remove tokens assert from dsv4 graph_mtp (#30444)

ORippler committed 16 hours ago

Verified 296538ad

model : allow Qwen3.8 Flash Next to accept image embds for MTP (#30397)

ORippler committed 16 hours ago

Verified 7f51953d

cuda : fix warnings from dead breaks (#30441)

angt committed 16 hours ago

Verified 882d6716

cuda : warnings cleanup (#30420)

angt committed 16 hours ago

Verified 30f7df1a

convert : fix LoRA conversion with --outtype auto (#30433)

teddygood committed 18 hours ago

Verified 9d7d5049

llama : skip logits buffer allocation when graph produces no logits (#30255)

animeshsri14 committed 18 hours ago

Verified ed019cd8

graph : avoid reallocations in the "no cache" path (#30439)

ggerganov committed 18 hours ago

Verified 98801578

ggml-zdnn : introduce Q8_0 mat mul (#29836)

taronaeo committed 18 hours ago

Verified af50f2f2

models : remove redundant ggml_cont calls now supported by all unary kernels (#30198)

saady789 committed 19 hours ago

Verified 071b553f

hexagon: extend scalar DMA broadcasting support for binary ops (#30245)

kurquhar committed 23 hours ago

Verified 4413a192

ci : build OpenVINO releases with GGML_NATIVE=OFF and CPU backend variants (#30324)

Pratyush-gg committed 1 day ago

Verified 8f574b0a

Older