ggerganov/llama.cpp

Pull Requests Commits

CUDA: make the alloc_deps check batch independent

am17an committed 6 days ago

9f80c69e

server: reject partial media truncation (#24076)

he-yufeng committed 6 days ago

Verified 8e164219

vulkan: fix stale prealloc_y reuse across flash attention and soft_max (#29591)

fxgsell committed 6 days ago

Verified 806eee98

vulkan: sparse flash attention for quantized K/V (#29639)

fxgsell committed 6 days ago

Verified b3daa077

llama : fix unexpected graph reallocation in the k-pool models (#29958)

ggerganov committed 6 days ago

Verified c173a53b

kv-cache: fix restoring mismatched KV cache rotation by saving exact rotation metadata (#28498)

eapache committed 6 days ago

Verified 21079106

ci : winget urls must be separate strings (#29978)

CISC committed 6 days ago

Verified e5983d67

ci : fix docker workflow permissions (#29979)

CISC committed 6 days ago

Verified 4ca6b76f

ggml-cpu : add Q8_0 IME1 matrix kernel for SpacemiT X60 (#28479)

alanhc committed 6 days ago

Verified 9f12cd4a

vulkan : Fix undeclared identifiers when -DGGML_VULKAN_RUN_TESTS=ON (#29912)

EAddario committed 6 days ago

Verified ebe18bee

webgpu: add MMVQ support for Q1_0/Q5_0/Q5_1/Q3_K/Q5_K/Q6_K/MXFP4 (#29483)

yomaytk committed 6 days ago

Verified 8216c846

cuda: tile the lightning indexer over keys and tokens for 4 heads (#29901)

ServeurpersoCom committed 6 days ago

Verified 1b43d311

metal : few-row MMA mat-mul (#29869)

pratiknarola-t committed 6 days ago

Verified a3a1c474

CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size (#29633)

ynankani committed 6 days ago

Verified 9d3aba6b

CUDA: prefer whole-tile FlashAttention scheduling for efficient two-stage kernels (#29435)

anujj committed 6 days ago

Verified d89651a7

log, server: self contained colors, split child commands from logs in router mode (#29895)

ServeurpersoCom committed 6 days ago

Verified a7fb71fa

llama: support both embd + raw tokens in batch (#29622)

ngxson committed 6 days ago

Verified 0bb496db

CUDA: refactor swizzling code (#29612)

JohannesGaessler committed 6 days ago

Verified 2ca15f54

ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (#29806)

SongXiaoXi committed 6 days ago

Verified a7b94df2

cuda : move neu_padded to where it is used (#29940)

angt committed 6 days ago

Verified 0eb6d9a8

ci : windows llvm build requires ninja multi-config (#29959)

CISC committed 6 days ago

Verified 2e7c58c5

ci : add windows arm64 vulkan release (#29954)

CISC committed 7 days ago

Verified 7f2dd88b

AGENTS.md : revamp (#29656)

am17an committed 7 days ago

Verified bf79dbbc

chat-peg-parser : clear current_tool when pending_tool_call is reset (#29942)

T-Anas committed 7 days ago

Verified dbe4c3ed

ci : set default permissions (#29945)

CISC committed 7 days ago

Verified 46847e61

cuda : move blocks_per_col to where it is used (#29939)

angt committed 7 days ago

Verified 2bc56357

CUDA: fix MMQ memory fault if n_expert >> n_ubatch (#29941)

JohannesGaessler committed 7 days ago

Verified dd266785

vulkan: fix rdna4 mat_vec tuning (#29934)

0cc4m committed 7 days ago

Verified 16c163d5

imatrix: calculate activation-based statistics for new format (GGUF) imatrices (#14891)

EAddario committed 7 days ago

Verified 05043961

spec : fix n-gram drafts rejected at temp > 0 after truncation (#29924)

praneshgo committed 7 days ago

Verified 8330e969

Older