ngxson/llama.cpp

Pull Requests Commits

ngxson committed 27 days ago

d0e14f73

Vulkan Flash Attention Coopmat1 Refactor (#19075)

0cc4m committed 27 days ago

Verified f6b533d8

spec : add self‑speculative decoding (no draft model required) + refactor (#18471)

srogmann committed 27 days ago

Verified 72d3b189

convert : yield Mamba2Model/GraniteMoeModel modify_tensors (#19157)

danbev committed 28 days ago

Verified ebf57258

ggml-sycl: remove unused syclcompat header (#19140)

PatKamin committed 28 days ago

Verified 0cd7032c

jinja : undefined should be treated as sequence/iterable (return string/array) by filters/tests (#19147)

CISC committed 28 days ago

Verified 60368e1d

vulkan: handle device dedup on MacOS + Vega II Duo cards (#19058)

okuvshynov committed 28 days ago

Verified 88d23ad5

doc: add build instruction to use Vulkan backend on macos (#19029)

chanbengz committed 28 days ago

Verified 0a95026d

ggml: new backend for Virglrenderer API Remoting acceleration (v2) (#18718)

kpouget committed 28 days ago

Verified b7feacf7

ggml-cpu: arm64: Q4_K scale unroll and vectorization (#19108)

Alcpz committed 28 days ago

Verified 6ad70c5a

cuda : fix "V is K view" check for non-unified KV cache (#19145)

ggerganov committed 28 days ago

Verified 631cbfcc

CUDA: tune GLM 4.7 Flash FA kernel selection logic (DGX Spark) (#19142)

ggerganov committed 28 days ago

Verified 2eee6c86

server : adjust spec tests to generate up to 16 tokens (#19093)

ggerganov committed 28 days ago

Verified b931f81b

llama : disable Direct IO by default (#19109)

ggerganov committed 28 days ago

Verified c5c64f72

sampling : remove sampling branching in output_reserve (#18811)

danbev committed 28 days ago

Verified eef375ce

ggml webgpu: Split shared state (webgpu_context) into global state and per-thread state (#18976)

nikhilJain17 committed 28 days ago

Verified 06961e28

ggml-zendnn : update ZenDNN git tag to main branch (#19133)

z-vishal committed 28 days ago

Verified f2571df8

jinja : implement mixed type object keys (#18955)

CISC committed 28 days ago

Verified 2b4cbd28

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

davelima committed 29 days ago

Verified 68ac3acb

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

JohannesGaessler committed 29 days ago

Verified a5bb8ba4

ci : revert slim runner for winget (#19129)

CISC committed 29 days ago

Verified c0204a08

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

Alcpz committed 29 days ago

Verified be8890e7

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

gaugarg-nv committed 29 days ago

Verified a83c73a1

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

danbev committed 29 days ago

Verified fc3cdf32

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

shalinib-ibm committed 29 days ago

Verified 7afdfc9b

opencl: add flattened q6_K mv (#19054)

lhez committed 29 days ago

Verified 94eeb596

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

JohannesGaessler committed 29 days ago

Verified b0311c16

graph : fix nkvo offload with FA (#19105)

ggerganov committed 29 days ago

Verified 8f80d1b2

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

CISC committed 30 days ago

Verified 142cbe2a

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

ggerganov committed 30 days ago

Verified 56f3ebf3

Older