ggerganov/llama.cpp

Pull Requests Commits

tests : add GQA=20 FA test

ggerganov committed 11 hours ago

Verified f0079c97

CUDA: faster FA for GQA > 1 but not power of 2 (#19092)

JohannesGaessler committed 12 hours ago

Verified 0c21677e

metal : fix recommendedMaxWorkingSetSize availability on legacy iOS/macOS (#19088)

ccbinn committed 14 hours ago

Verified 0440bfd1

convert : yield Gemma3N custom_map tensors directly (#19091)

CISC committed 15 hours ago

Verified 0bf56369

ggml-cpu: Use tiled FA for prompt-processing (#19012)

am17an committed 17 hours ago

Verified bcb43163

kv-cache : support V-less cache (#19067)

ggerganov committed 18 hours ago

Verified d9c6ce46

convert : fix Gemma3N, GraniteMoe and Ernie4.5Moe (#19084)

CISC committed 20 hours ago

Verified 70d86082

completion : fix prompt cache for recurrent models (#19045)

ggerganov committed 1 day ago

Verified 080b1619

readme: update RWKV7 model links (#19061)

MollySophia committed 1 day ago

Verified 1243f93a

llama: fix integer type consistency in split helpers (#18894)

MaheshJakkala committed 1 day ago

Verified 24bc2383

common : use two decimal places for float arg help messages (#19048)

danbev committed 1 day ago

Verified 16639ba2

convert : fix conversion for inheriting models that were bypassing modify_tensors (#19064)

bartowski1182 committed 1 day ago

Verified 9981c301

llama-fit-params: keep explicit --ctx-size 0 (#19070)

JohannesGaessler committed 1 day ago

Verified e9fd8dca

GGUF: check that tensor size is representable (#19072)

JohannesGaessler committed 1 day ago

Verified 4e5b83b2

chat: fix language input for translategemma (#19052)

ngxson committed 1 day ago

Verified bb02f74c

CUDA: re-use MLA K data for V in MMA FA (#19057)

JohannesGaessler committed 1 day ago

Verified 8f91ca54

ggml-cuda: enable cuda-graphs for `n-cpu-moe` (#18934)

am17an committed 2 days ago

Verified 81ab64f3

ggml-hexagon: flash-attn opt (#19025)

chraac committed 2 days ago

Verified 8af1f5f4

graph : utilize `ggml_build_forward_select()` to avoid reallocations (#18898)

ggerganov committed 2 days ago

Verified 557515be

[SYCL] use malloc to support both iGPU and dGPU in same time (#18992)

arthw committed 2 days ago

Verified cb6caca1

chat : fix translategemma crash on common_chat_format_example (#19019)

ngxson committed 2 days ago

Verified b5b8fa1c

model-conversion : use BUILD_DIR variable in all scripts (#19015)

danbev committed 3 days ago

Verified a14b960b

ggml-cpu: aarm64: q5_K repack gemm and gemv (and generic) implementations (i8mm) (#18860)

Alcpz committed 3 days ago

Verified 091a46cb

cli : load parser definition (#19031)

aldehir committed 3 days ago

Verified a3e81281

server : support preserving reasoning_content in assistant message (#18994)

ngxson committed 3 days ago

Verified 51fa458a

mla : make the V tensor a view of K (#18986)

ggerganov committed 3 days ago

Verified a5eaa1d6

CUDA: fix alignment check for FA (#19023)

JohannesGaessler committed 3 days ago

Verified e2baf021

convert_hf_to_gguf.py: refactor modify_tensors to call super (#18866)

am17an committed 3 days ago

Verified e34d6d03

opencl: enable the general fp mm for non-cont input and as a fallback for specialized kqv kernel for adreno (#18970)

lhez committed 3 days ago

Verified 9c96465f

server: do not log certain endpoints (avoid log spam) (#19028)

ngxson committed 3 days ago

Verified 4e595b25

Older