ggml-org/llama.cpp

Pull Requests Commits

llama : bump max seq limit from 64 to 256

ggerganov committed 20 hours ago

Verified 333c9ead

CANN: Add ROPE sin/cos cache for reuse (#15912)

noemotiovon committed 20 hours ago

Verified 10d8b2b6

CANN: implement LRU cache for ACL graphs (#15814)

noemotiovon committed 1 day ago

Verified 28b5f190

llama : check returned fn ptrs from ggml_backend_reg_get_proc_address (#15893)

danbev committed 1 day ago

Verified 86587da0

ci : cache ROCm installation in windows-latest-cmake-hip (#15887)

danbev committed 1 day ago

Verified ff02caf9

vulkan: throw the oom error instead of no memory type found (#15905)

0cc4m committed 1 day ago

Verified ae355f6f

vulkan: Fix OOB accesses in soft_max_back (#15861)

jeffbolznv committed 1 day ago

Verified 4f63cd70

HIP: use v_dot2_f32_f16 instruction for FA (#15884)

JohannesGaessler committed 1 day ago

Verified 17bc5a81

Workaround for subgroup arithmetic failing on MoltenVK with AMD GPUs (issue 15846) (#15886)

lksj92hs committed 1 day ago

Verified ed54e325

CUDA: Add mul_mat_id support for the mmf kernel (#15767)

am17an committed 2 days ago

Verified a972faeb

CUDA: fix GET_ROWS for large tensors (#15882)

JohannesGaessler committed 2 days ago

Verified 550cf726

contrib : add notes about merging PRs (#15881)

ggerganov committed 2 days ago

Verified c252ce67

requirements : update transformers/torch for Embedding Gemma (#15828)

danbev committed 2 days ago

Verified 70cd37db

model-conversion : add extra debugging support for model conversion (#15877)

pwilkin committed 2 days ago

Verified acc1b008

json : support `enum` values within `allOf` (#15830)

aldehir committed 2 days ago

Verified 7057faf6

media : add llama1 icon (#15878)

06kellyjac committed 2 days ago

Verified fe1c92cd

vulkan: sort graph to allow more parallel execution (#15850)

jeffbolznv committed 2 days ago

Verified e68aa10d

CUDA: generate_cu_files.py - add missing mxfp4 (#15880)

am17an committed 2 days ago

Verified 0a16bf52

chat : Deepseek V3.1 reasoning and tool calling support (OpenAI Style) (#15533)

createthis committed 2 days ago

Verified 88021565

server : bring back timings_per_token (#15879)

ngxson committed 2 days ago

Verified 56920f56

cuda : fix supports_op condition for get_rows when number of blocks is too large (#15868)

ggerganov committed 2 days ago

Verified b0d52998

metal : refactor + optimize (#15857)

ggerganov committed 2 days ago

Verified f28d4f4a

ggml: allow casting between f32 and i32 (#15783)

ngxson committed 2 days ago

Verified 9fcb29f2

CUDA: non-contiguous src0 not supported for PAD (#15869)

CISC committed 2 days ago

Verified 5ef22d28

convert : force setting sliding_window from original config (#15867)

danbev committed 2 days ago

Verified 233d773d

batched-bench : fix llama_synchronize usage during prompt processing (#15835)

ggerganov committed 3 days ago

Verified a885dcff

context : fix n_outputs during reserve (#15858)

ggerganov committed 3 days ago

Verified 663027fd

model : avoid ggml_cont_3d for fused QKV weights (#15662)

ggerganov committed 3 days ago

Verified cf0e3ba1

tests: large sizes for get_rows (#15687)

jeffbolznv committed 3 days ago

Verified d413dca0

CANN: Stream sync between devices for acl_graph (#15809)

noemotiovon committed 3 days ago

Verified 85ca66a7

Older