ggerganov/llama.cpp

Pull Requests Commits

ggml-quants : handle zero amax for MXFP4

compilade committed 28 days ago

2763dc8b

gguf-py : add MXFP4 de/quantization support

compilade committed 29 days ago

141cab13

llama : add gpt-oss (#15091)

ggerganov committed 30 days ago

Verified fd1234cb

chat : only remove double bos/eos if added (#15086)

CISC committed 30 days ago

Verified f324a3b7

readme : update hot topics (#15097)

ggerganov committed 30 days ago

Verified be426425

sycl: fix mul_mat selection (#15092)

Rbiessy committed 30 days ago

Verified 3306ceab

Fix `glm4moe` bug (#15088)

jukofyork committed 30 days ago

Verified c81de6e1

webui: fix markdown table (#15081)

dindinw committed 30 days ago

Verified 22f060c9

context : fix index overflow on huge outputs (#15080)

compilade committed 30 days ago

Verified ee3a9fcf

llama : add --n-cpu-moe option (#15077)

slaren committed 30 days ago

Verified ec428b02

imatrix : warn when GGUF imatrix is saved without .gguf suffix (#15076)

compilade committed 30 days ago

Verified 19f68fa5

cmake: Add GGML_BACKEND_DIR option (#15074)

ckastner committed 30 days ago

Verified 41613437

gguf-py : add --chat-template-file to gguf_new_metadata (#15075)

CISC committed 31 days ago

Verified e5bebe52

model: support GLM 4.5 family of models (#14939)

sammcj committed 31 days ago

Verified ef0144c0

quantize : fix confusing error message if ftype is invalid (#15071)

CISC committed 31 days ago

Verified 2721257e

ggml: WebGPU backend host improvements and style fixing (#14978)

reeselevine committed 31 days ago

Verified 587d0118

vulkan: fix build when using glslang that does not support coopmat2 (#15062)

jeffbolznv committed 31 days ago

Verified 5aa1105d

imatrix : use GGUF by default (#14842)

compilade committed 31 days ago

Verified d31192b4

imatrix : fix 3d activation handling for hybrid and recurrent models (#14994)

compilade committed 31 days ago

Verified 0a2f5496

memory : handle kv_unified for hybrid models (#15050)

compilade committed 31 days ago

Verified 11a38111

vocab : JetBrains Mellum pre-tokenizer (#15045)

csabakecskemeti committed 31 days ago

Verified 97366dc6

model : add text-only support for Kimi-VL (and find special tokens in text_config) (#15051)

gabriellarson committed 32 days ago

Verified 83bc2f28

vulkan: Use coopmat2 for conv2d (#14982)

jeffbolznv committed 32 days ago

Verified 6c7a4411

opencl: fix adreno compiler detection logic (#15029)

lhez committed 33 days ago

Verified 5c0eb5ef

CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)

JohannesGaessler committed 33 days ago

Verified 03d46982

cuda: make im2col a little faster (#15025)

leejet committed 33 days ago

Verified 3303c19b

kv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)

danbev committed 33 days ago

Verified 4fdea540

llama : enable LLAMA_SET_ROWS=1 by default (#14959)

ggerganov committed 33 days ago

Verified a4569c41

cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (#15038)

ggerganov committed 33 days ago

Verified 15e92fd3

ci : check that pre-tokenizer hashes are up-to-date (#15032)

CISC committed 33 days ago

Verified 2bf3fbf0

Older