ggerganov/llama.cpp

Pull Requests Commits

move set_rows to n_task = n_threads group

am17an committed 26 days ago

34eca226

ggml-cpu : size set_rows conversion buffer for all threads

am17an committed 26 days ago

3467657f

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

ggerganov committed 26 days ago

Verified a17e94d2

llama : fix log message (#30452)

ggerganov committed 26 days ago

Verified b476d9a7

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

0cc4m committed 26 days ago

Verified e0bcaccb

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

homeofe committed 26 days ago

Verified 0a731cce

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

ggerganov committed 26 days ago

Verified f851e244

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

angt committed 26 days ago

Verified bfe65e1d

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

losewayy committed 26 days ago

Verified 411d5485

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

ORippler committed 26 days ago

Verified 296538ad

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

ORippler committed 26 days ago

Verified 7f51953d

cuda : fix warnings from dead breaks (#30441)

angt committed 26 days ago

Verified 882d6716

cuda : warnings cleanup (#30420)

angt committed 27 days ago

Verified 30f7df1a

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

teddygood committed 27 days ago

Verified 9d7d5049

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

animeshsri14 committed 27 days ago

Verified ed019cd8

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

ggerganov committed 27 days ago

Verified 98801578

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

taronaeo committed 27 days ago

Verified af50f2f2

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

saady789 committed 27 days ago

Verified 071b553f

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

kurquhar committed 27 days ago

Verified 4413a192

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

Pratyush-gg committed 27 days ago

Verified 8f574b0a

ci: fix OpenVINO download links (#30401)

Pratyush-gg committed 27 days ago

Verified a49f58dc

model: fix vllm d3 multi seq case (#30408)

ngxson committed 27 days ago

Verified cbe28bb3

graph: pad the KV dimension of the no-cache attention to 256 (#30416)

ServeurpersoCom committed 27 days ago

Verified 358a420e

metal: fix metal tensor kernel_mul_mm error for large tensor (#28748)

SecondSeed committed 27 days ago

Verified 48e98575

hexagon: optimize CPY/CONCAT — round 3, short-rows edition (#30341)

max-krasnyansky committed 27 days ago

Verified 8bb3cf0d

vulkan: pack FMA for f16 accumulation, no dot product op (#29877)

amitmi704 committed 27 days ago

Verified 0f701c88

model: support vllm d3 models (#30383)

ngxson committed 27 days ago

Verified efe3626b

model: support vllm decision 2 model (#30378)

ngxson committed 27 days ago

Verified 097cf33d

tests : link each test against only the libraries it uses (#30299)

ggerganov committed 27 days ago

Verified ccdbe290

metal : fix mul_mat_id with duplicate ids (#30373)

ggerganov committed 27 days ago

Verified 073b7403

Older