ggml-org/llama.cpp

Pull Requests Commits

CUDA: tune GLM 4.7 Flash FA kernel selection logic (DGX Spark)

ggerganov committed 6 hours ago

f4837d3e

docs: Remove duplicated word on CUDA build section (#19136)

davelima committed 9 hours ago

Verified 68ac3acb

CUDA: tune GLM 4.7 Flash FA kernel selection logic (#19097)

JohannesGaessler committed 9 hours ago

Verified a5bb8ba4

ci : revert slim runner for winget (#19129)

CISC committed 12 hours ago

Verified c0204a08

ggml-cpu: aarm64: q6_K repack gemm and gemv (and generic) implementations (i8mm) #18860 (#18888)

Alcpz committed 14 hours ago

Verified be8890e7

[CUDA] Reduce CPU-side stalls due to the CUDA command buffer being full (#19042)

gaugarg-nv committed 16 hours ago

Verified a83c73a1

common : clarify HTTPS build options in error message (#19103)

danbev committed 17 hours ago

Verified fc3cdf32

ggml-cpu: Enable FP16 MMA kernels on PPC (#19060)

shalinib-ibm committed 19 hours ago

Verified 7afdfc9b

opencl: add flattened q6_K mv (#19054)

lhez committed 19 hours ago

Verified 94eeb596

CUDA: fix padding of GQA to power of 2 in FA (#19115)

JohannesGaessler committed 1 day ago

Verified b0311c16

graph : fix nkvo offload with FA (#19105)

ggerganov committed 1 day ago

Verified 8f80d1b2

ci : use new 1vCPU runner for lightweight jobs (#19107)

CISC committed 1 day ago

Verified 142cbe2a

model : add correct type for GLM 4.7 Flash (#19106)

ggerganov committed 1 day ago

Verified 56f3ebf3

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

JohannesGaessler committed 2 days ago

Verified 0c21677e

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

ccbinn committed 2 days ago

Verified 0440bfd1

convert : yield Gemma3N custom_map tensors directly (#19091)

CISC committed 2 days ago

Verified 0bf56369

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

am17an committed 2 days ago

Verified bcb43163

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

ggerganov committed 2 days ago

Verified d9c6ce46

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

CISC committed 2 days ago

Verified 70d86082

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

ggerganov committed 2 days ago

Verified 080b1619

readme: update RWKV7 model links (#19061)

MollySophia committed 2 days ago

Verified 1243f93a

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

MaheshJakkala committed 2 days ago

Verified 24bc2383

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

danbev committed 2 days ago

Verified 16639ba2

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

bartowski1182 committed 2 days ago

Verified 9981c301

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

JohannesGaessler committed 3 days ago

Verified e9fd8dca

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

JohannesGaessler committed 3 days ago

Verified 4e5b83b2

chat: fix language input for translategemma (#19052)

ngxson committed 3 days ago

Verified bb02f74c

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

JohannesGaessler committed 3 days ago

Verified 8f91ca54

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

am17an committed 3 days ago

Verified 81ab64f3

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

chraac committed 3 days ago

Verified 8af1f5f4

Older