ggerganov/llama.cpp

Pull Requests Commits

ggml: allow input async copies

am17an committed 6 days ago

4815e5e7

HIP: use 64-row MMQ tiles for some quant types on RDNA3.5 (#28195)

liangliangchang committed 8 days ago

Verified 4dd6dd4b

vulkan: fix compile error (#30495)

0cc4m committed 8 days ago

Verified 14310190

llama : build the k-pool selection in the causal mask layout (#30480)

ggerganov committed 8 days ago

Verified e7268bf6

vulkan: Add workaround for device free memory > total memory on Intel Windows (#29835)

rillomas committed 8 days ago

Verified cdce056a

vulkan: Define the shader count for Intel Arc Pro B65 + B70 (#30431)

TheBlueMatt committed 8 days ago

Verified d524bb66

vulkan: support mulmat bf16 activations (#29069)

0cc4m committed 8 days ago

Verified 7ec420bc

unicode : add opt-in NFC normalization for Unigram tokenizers (#30390)

sandrohanea committed 8 days ago

Verified 10da8c46

convert : export YaRN truncation config (#30288)

tokinasin committed 8 days ago

Verified eca5458c

Enable OpenSSL in FP32/FP16 SYCL builds to restore HF HTTPS downloads (#30330)

raffaeleedidonna committed 8 days ago

Verified 091b6a7f

webgpu: (Fix CI) fix mul_mat_id with duplicate expert ids (#30451)

yomaytk committed 8 days ago

Verified fcac1907

server : escape the special tokens without std::regex (#30243)

ykhrustalev committed 8 days ago

Verified 04f7226a

chat: replace output string with parts (#30389)

ngxson committed 8 days ago

Verified ff026f8c

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

angt committed 9 days ago

Verified 1b70350c

docs : update zDNN supported ops (#30472)

taronaeo committed 9 days ago

Verified d71e324a

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

chaxu01 committed 9 days ago

Verified 45451c36

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

jeffbolznv committed 9 days ago

Verified 77c4540c

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

Xanewok committed 9 days ago

Verified ecdd29e8

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

pdhinaka committed 9 days ago

Verified ec0bfac1

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

ServeurpersoCom committed 9 days ago

Verified ace20f67

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

ggerganov committed 9 days ago

Verified b9f3599f

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

accaldwell committed 9 days ago

Verified f441d2d9

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

ServeurpersoCom committed 9 days ago

Verified cb55ffe6

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

ggerganov committed 9 days ago

Verified a17e94d2

llama : fix log message (#30452)

ggerganov committed 9 days ago

Verified b476d9a7

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

0cc4m committed 9 days ago

Verified e0bcaccb

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

homeofe committed 9 days ago

Verified 0a731cce

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

ggerganov committed 9 days ago

Verified f851e244

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

angt committed 9 days ago

Verified bfe65e1d

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

losewayy committed 9 days ago

Verified 411d5485

Older