ggml-org/llama.cpp

Pull Requests Commits

convert : for FP8, use scale type to decide auto type

compilade committed 36 days ago

2ef41855

gguf-py : handle cross-filesystem file range copies

compilade committed 36 days ago

f88a4b93

convert : better logging of partially reflinkable tensors

compilade committed 36 days ago

4be1a5d4

gguf-py : allow previewing reflinked size on non-Linux platforms

compilade committed 36 days ago

6ffa46d8

convert : remove unused field ModelTensorInfo.src_qtype

compilade committed 36 days ago

3126b5ee

convert : more robust default ftype detection

compilade committed 36 days ago

e097d98a

gguf-py : improve reflink size logging

compilade committed 36 days ago

5712aa89

convert : allow sharding reflinked models

compilade committed 36 days ago

d3fcb0e9

convert : use F32 operations on Mamba A_log

compilade committed 36 days ago

614b95a8

convert : detect filesystem block size for reflinks

compilade committed 36 days ago

c3738cfc

gguf-py : fix flake8 lint

compilade committed 36 days ago

791bd97b

convert : fix reflinks for stacked MoE tensors

compilade committed 36 days ago

d9210570

convert : use reflinks for faster conversion

compilade committed 36 days ago

562aa42c

convert : fix no-lazy dtypes from direct safetensors

compilade committed 36 days ago

e996f3ae

gguf-py : order safetensors tensors by name

compilade committed 36 days ago

e7b7ed8a

convert : parse safetensors directly

compilade committed 36 days ago

c4b630f2

ggml-cpu : optimize RVV q2_k and q3_k kernels (#16887)

xctan committed 37 days ago

Verified 7f09a680

CUDA: fix crash on uneven context without FA (#16988)

JohannesGaessler committed 37 days ago

Verified aa374175

metal : initial Metal4 tensor API support (#16634)

ggerganov committed 37 days ago

Verified 5b180c3d

server : disable checkpoints with mtmd (#17045)

ggerganov committed 37 days ago

Verified b7f9010d

clip: implement minicpm-v sinusoidal embd using GGML (#17036)

ngxson committed 37 days ago

Verified 4882f0ff

sycl: add CONCAT operator support (#16047)

ye-NX committed 37 days ago

Verified 9d7c518d

docs: explain CUDA 11 compilation [no ci] (#16824)

JohannesGaessler committed 37 days ago

Verified 22c8c3c6

ggml-hexagon: graceful fallback for older socs where rpcmem_alloc2 and FASTRPC_GET_URI is unsupported (#16987)

l3utterfly committed 37 days ago

Verified 6db3d1ff

improve CUDA cpy memory bandwidth when copying transposed tensor (#16841)

bssrdf committed 38 days ago

Verified 230d1169

vulkan: Fix GGML_VULKAN_CHECK_RESULTS to better handle fusion (#16919)

jeffbolznv committed 38 days ago

Verified a44d7712

examples(gguf): GGUF example outputs (#17025)

gabe-l-hart committed 38 days ago

Verified 5886f4f5

mtmd: allow QwenVL to process larger image by default (#17020)

ngxson committed 38 days ago

Verified 92bb84f7

server : do not default to multiple slots with speculative decoding (#17017)

ggerganov committed 38 days ago

Verified 13b339bc

mtmd: improve struct initialization (#16981)

ngxson committed 38 days ago

Verified 2f0c2db4

Older