Commits

Commit history

1. Commits on Oct 16, 2026

   • hexagon: vectorize softplus with HVX (#30435)

     Show description for 27c7a76
     huangwei858

     authored
     27c7a76View commit details

     Copy full SHA for 27c7a76

     Browse repository at this point

   • opencl: Add MxFP4 MoE bin kernel for Adreno X1 devices (#30691)

     Show description for 2f2763d
     

     shawngu-quic

     and

     lhez

     authored
     2f2763dView commit details

     Copy full SHA for 2f2763d

     Browse repository at this point

   • model: add Beam (Reflection) (#30670)

     danielhanchen

     authored
     01fc3aaView commit details

     Copy full SHA for 01fc3aa

     Browse repository at this point

   • model : Granite Switch Decision (#30676)

     Show description for 23f7aab
     gabe-l-hart

     authored
     23f7aabView commit details

     Copy full SHA for 23f7aab

     Browse repository at this point

   • common: honor HTTP_PROXY/HTTPS_PROXY env vars in http client (#21752)

     Show description for 7182ef7
     

     3 peopleauthored
     7182ef7View commit details

     Copy full SHA for 7182ef7

     Browse repository at this point

   • convert : respect Mistral format for DeepseekV2Model (#30669)

     Show description for 2a66703
     aiqubits

     authored
     2a66703View commit details

     Copy full SHA for 2a66703

     Browse repository at this point

   • Add more benchmarks to llama-eval (#26174)

     Show description for f336359
     pwilkin

     authored
     f336359View commit details

     Copy full SHA for f336359

     Browse repository at this point

   • hexagon: correct HMX FA sinks (#30581)

     Show description for ceefd16
     

     jhen0409

     and

     max-krasnyansky

     authored
     ceefd16View commit details

     Copy full SHA for ceefd16

     Browse repository at this point

   • model : MTP support for Mellum2.1 (#30612)

     Show description for 9a16f1d
     Xarbirus

     authored
     9a16f1dView commit details

     Copy full SHA for 9a16f1d

     Browse repository at this point

   • convert : speed up model endian conversion and add more datatype support (#30647)

     Show description for b84ea5c
     taronaeo

     authored
     b84ea5cView commit details

     Copy full SHA for b84ea5c

     Browse repository at this point

   • [SYCL]: fix memory errors in mul_mat, split buffer, host pool (#29889)

     mctylr-gh

     authored
     a30fb49View commit details

     Copy full SHA for a30fb49

     Browse repository at this point

   • gguf-py : add numpy Q1_0 and Q2_0 de/quantization support (#30630)

     Show description for 495c580
     SomeNobody21112

     authored
     495c580View commit details

     Copy full SHA for 495c580

     Browse repository at this point

   • CUDA: HIP: rms_norm: add variants that load float4 at a time and store x in registers (#29720)

     Show description for 623c163
     IMbackK

     authored
     623c163View commit details

     Copy full SHA for 623c163

     Browse repository at this point

   • common : read ngram cache parts by reference instead of copying (#30061)

     jadidbourbaki

     authored
     13b8048View commit details

     Copy full SHA for 13b8048

     Browse repository at this point

   • convert : support text-only checkpoints in pplx-decider (#30575)

     Show description for 0f43762
     final-bench

     authored
     0f43762View commit details

     Copy full SHA for 0f43762

     Browse repository at this point

   • cpu : reduce AVX2 tiled accumulators together (#30296)

     Show description for 3dbcb7f
     trivalik

     authored
     3dbcb7fView commit details

     Copy full SHA for 3dbcb7f

     Browse repository at this point

   • ggml-cpu: refactor s309x repack helper functions (#30588)

     Show description for c804754
     taronaeo

     authored
     c804754View commit details

     Copy full SHA for c804754

     Browse repository at this point

   • sycl : reduce the build threads to half (#30236)

     arthw

     authored
     c0083e3View commit details

     Copy full SHA for c0083e3

     Browse repository at this point

   • sycl: wide stores for the q4_K/q5_K weight dequant (#29696)

     Titaniumtown

     authored
     13de11eView commit details

     Copy full SHA for 13de11e

     Browse repository at this point

   • tests : catch vanished rows in the fusion baseline (#30618)

     Show description for 88aaf32
     ggerganov

     authored
     88aaf32View commit details

     Copy full SHA for 88aaf32

     Browse repository at this point

   • vulkan: fix gpu-visible host memory issues (#30468)

     Show description for a8d9a7e
     0cc4m

     authored
     a8d9a7eView commit details

     Copy full SHA for a8d9a7e

     Browse repository at this point

   • Fix failing RocM tests (#30611)

     Show description for f12aaa3
     

     mailkv23

     and

     Karthik Vijayan

     authored
     f12aaa3View commit details

     Copy full SHA for f12aaa3

     Browse repository at this point

   • graph : add precision handling for MELLUM architecture (#30584)

     Xarbirus

     authored
     5e03a1fView commit details

     Copy full SHA for 5e03a1f

     Browse repository at this point

   • hexagon: tensor repacking rework and support for tiled and flat quantized and fp weights (#30635)

     Show description for b797b9c
     max-krasnyansky

     authored
     b797b9cView commit details

     Copy full SHA for b797b9c

     Browse repository at this point

   • llama : share the k-pool input views between layers (#30613)

     Show description for 6afa9a3
     am17an

     authored
     6afa9a3View commit details

     Copy full SHA for 6afa9a3

     Browse repository at this point

   • Limit max CUDA pinned host buffer size to 4 GB (#30449)

     gaugarg-nv

     authored
     b6334e3View commit details

     Copy full SHA for b6334e3

     Browse repository at this point

   • sycl: support reordered Q8_0 MoE weights (#29337)

     Show description for e707790
     newjordan

     authored
     e707790View commit details

     Copy full SHA for e707790

     Browse repository at this point

   • [SYCL] Enable ssl in build script and CI (#30456)

     Show description for 43d57fb
     

     arthw

     and

     CISC

     authored
     43d57fbView commit details

     Copy full SHA for 43d57fb

     Browse repository at this point
2. Commits on Oct 15, 2026

   • mtmd: add mtmd_audio_stream + support voxtral-realtime (#30629)

     Show description for 2851b3b
     ngxson

     authored
     2851b3bView commit details

     Copy full SHA for 2851b3b

     Browse repository at this point

   • convert : fix Granite Switch block_count for MultiSwitch checkpoints (#30621)

     Show description for 59d5fad
     

     planetf1

     and

     gabe-l-hart

     authored
     59d5fadView commit details

     Copy full SHA for 59d5fad

     Browse repository at this point

   • convert : fix regression in mapping of Granite speech tensors (#30628)

     Show description for e287bd8
     mrutkows

     authored
     e287bd8View commit details

     Copy full SHA for e287bd8

     Browse repository at this point

   • mtmd: lightonocr does not require longest edge to always be 1540, native resolution is the default (#30600)

     Show description for 2f729cf
     staghado

     authored
     2f729cfView commit details

     Copy full SHA for 2f729cf

     Browse repository at this point

   • model: add Muse Spark Open (text, vision and audio) (#30616)

     Show description for ba3c05f
     

     4 peopleauthored
     ba3c05fView commit details

     Copy full SHA for ba3c05f

     Browse repository at this point

   • pi : expand AI disclosure with hardware and stats (#30615)

     Show description for b76b133
     ggerganov

     authored
     b76b133View commit details

     Copy full SHA for b76b133

     Browse repository at this point

   • llama.cpp : bump version to 0.7.0 (#30603)

     ggerganov

     authored
     620d222View commit details

     Copy full SHA for 620d222

     Browse repository at this point