Commits

Commits on Oct 18, 2025

opencl: transposed gemm/gemv moe kernel with mxfp4,f32 (#16602)
Show description for 8138785
shawngu-quic
authored
8138785
Copy full SHA for 8138785

Commits on Oct 17, 2025

llama-model: fix insonsistent ctxs <-> bufs order (#16581)
JohannesGaessler
authored
66b0dbc
Copy full SHA for 66b0dbc
rpc : report actual free memory (#16616)
Show description for 41386cf
rgerganov
authored
41386cf
Copy full SHA for 41386cf
vulkan: Add State Space Model (SSM) Operations Support (#16463)
Show description for 3d4e86b
giuseppe
authored
3d4e86b
Copy full SHA for 3d4e86b
ggml : fix SpaceMit IME array out-of-bounds in task assignment (#16629)
Show description for 342c728

muggle-stack
and
muggle
authored
342c728
Copy full SHA for 342c728
webui: reorganize settings layout (#16607)
Show description for ababae7
ServeurpersoCom
authored
ababae7
Copy full SHA for ababae7
vulkan: fix debug build (add_rms_len/data not found) (#16624)
jeffbolznv
authored
b194915
Copy full SHA for b194915
metal : add `CONV_TRANSPOSE_2D` (#16542)
Show description for 9ad4f19

iliailmer
and
ggerganov
authored
9ad4f19
Copy full SHA for 9ad4f19
grammar : use int64_t to avoid int overflows in int schema to grammar conversion logic (#16626)
ochafik
authored
79967ec
Copy full SHA for 79967ec
SYCL SET operator optimized for F32 tensors (#16350)
Show description for ceff6bb

GittyBurstein
and
Gitty Burstein
authored
ceff6bb
Copy full SHA for ceff6bb

Commits on Oct 16, 2025

mtmd : support home-cooked Mistral Small Omni (#14928)
ngxson
authored
1bb4f43
Copy full SHA for 1bb4f43
fix: added a normalization step for MathJax-style \[\] and \(\) delimiters (#16599)
Show description for 683fa6b
ServeurpersoCom
authored
683fa6b
Copy full SHA for 683fa6b
sycl : add ARANGE operator (#16362)
Show description for b22572e

GittyBurstein
and
Gitty Burstein
authored
b22572e
Copy full SHA for b22572e
CANN: format code using .clang-format (#15863)
Show description for 7a50cf3

noemotiovon
and
hipudding
authored
7a50cf3
Copy full SHA for 7a50cf3
common : Update the docs on -t --threads (#16236)
Show description for 6f5d924
takasurazeem
authored
6f5d924
Copy full SHA for 6f5d924
ggml-cpu: replace putenv with setenv for const-correctness (#16573)
Show description for adc9b60
otegami
authored
adc9b60
Copy full SHA for adc9b60
SYCL: Add GGML_OP_MEAN operator support (#16009)
Show description for ee50ee1

yael-works
and
CISC
authored
ee50ee1
Copy full SHA for ee50ee1

Commits on Oct 15, 2025

gguf-py : add support for endian conversion of BF16 data (#16594)
Show description for 7adc79c
AlekseiNikiforovIBM
authored
7adc79c
Copy full SHA for 7adc79c
cpu : add FLOOR, CEIL, ROUND and TRUNC unary operators (#16083)
Show description for 466c191
safranowith
authored
466c191
Copy full SHA for 466c191
opencl: add q8_0 mm support (#16469)
Show description for 0cb7a06
lhez
authored
0cb7a06
Copy full SHA for 0cb7a06
opencl: fix FA for f32 (#16584)
lhez
authored
d93f843
Copy full SHA for d93f843
Add server-driven parameter defaults and syncing (#16515)
allozaur
authored
f9fb33f
Copy full SHA for f9fb33f
metal: optimise `GGML_OP_SUM` (#16559)
Show description for f4ce81c

cern1710
and
ggerganov
authored
f4ce81c
Copy full SHA for f4ce81c
server : fix img token logs (#16595)
ggerganov
authored
17304cb
Copy full SHA for 17304cb
llama-quant: add support for mmproj (#16592)
Show description for 3e3cb19

ngxson
and
ggerganov
authored
3e3cb19
Copy full SHA for 3e3cb19
CUDA: Changing the CUDA scheduling strategy to spin (#16585)
Show description for 5acd455

JTischbein
and
JohannesGaessler
authored
5acd455
Copy full SHA for 5acd455
server : fix mtmd checkpoints (#16591)
ggerganov
authored
554fd57
Copy full SHA for 554fd57

Commits on Oct 14, 2025

metal : avoid using Metal's gpuAddress property (#16576)
Show description for fa882fd
ggerganov
authored
fa882fd
Copy full SHA for fa882fd
vulkan: Add ACC_TYPE_VEC2 implementation (#16203)
Show description for ffa0590

SavicStefan
and
Stefan Savic
authored
ffa0590
Copy full SHA for ffa0590
CUDA + openCL: fix bug in accessing rms_norm->src while doing fusion (#16577)
am17an
authored
120bf70
Copy full SHA for 120bf70
vulkan: Support FA with K/V in F32 (#16543)
jeffbolznv
authored
4258e0c
Copy full SHA for 4258e0c
vulkan: Improve build time for MSVC (#16545)
Show description for 7ea15bb
jeffbolznv
authored
7ea15bb
Copy full SHA for 7ea15bb
CUDA: enable FA for FP32 KV cache (#16546)
JohannesGaessler
authored
9c7185d
Copy full SHA for 9c7185d
CUDA: use fastdiv + ggml_cuda_mad for mmvf (#16557)
Show description for 1ee9d0b
am17an
authored
1ee9d0b
Copy full SHA for 1ee9d0b
CUDA: add fp kernel for larger batch size MoE (#16512)
Show description for 48e2fa9
am17an
authored
48e2fa9
Copy full SHA for 48e2fa9