Commits

Branch selector

master

User selector

All users

All time

Commit history

Commits on Jul 19, 2026

• Merge remote-tracking branch 'upstream/master'

  gary149

  committed
  f473a1aView commit details

  Copy full SHA for f473a1a

  Browse repository at this point

Commits on Jul 18, 2026

• model: rotate injected K/V cache for DFlash (#25823)

  Show description for 571d0d5
  

  ruixiang63

  and

  ggerganov

  authored
  571d0d5View commit details

  Copy full SHA for 571d0d5

  Browse repository at this point

• llama-quant : exclude i32 ffn_gate_tid2eid routing table from quantization (#25787)

  Show description for 4937ca8
  devYRPauli

  authored
  4937ca8View commit details

  Copy full SHA for 4937ca8

  Browse repository at this point

Commits on Jul 17, 2026

• opencl: load and use `kernel_gemm_moe_q6_k_f32_ns` from bin kernel lib (#25797)

  lhez

  authored
  86a9c79View commit details

  Copy full SHA for 86a9c79

  Browse repository at this point

• opencl: read/write MoE dp4a activation tiles to local memory as 128-bit (vectorized LD/ST perf opt) for Adreno GPUs (#25810)

  Show description for 6bdd77f
  wanghqc

  authored
  6bdd77fView commit details

  Copy full SHA for 6bdd77f

  Browse repository at this point

• opencl: transpose q4_K noshuffle scales for coalesced reads (#25805)

  wanghqc

  authored
  86d86edView commit details

  Copy full SHA for 86d86ed

  Browse repository at this point

• sync : ggml

  ggerganov

  committed
  7d56da7View commit details

  Copy full SHA for 7d56da7

  Browse repository at this point

• ggml : bump version to 0.17.0 (ggml/1568)

  ggerganov

  committed
  3727404View commit details

  Copy full SHA for 3727404

  Browse repository at this point

• tests : initialize all tensors in test_dsv4_hc to avoid NaNs in sentinel tensors (#25822)

  Show description for 5d5306b
  

  fairydreaming

  and

  sszymczy

  authored
  5d5306bView commit details

  Copy full SHA for 5d5306b

  Browse repository at this point

• common : auto-download dflash- and eagle3- HF sidecars (#25811)

  Show description for 635cdd5
  ggerganov

  authored
  635cdd5View commit details

  Copy full SHA for 635cdd5

  Browse repository at this point

• ggml-blas: default hadamard mul_mat to cpu routine (#25710)

  Show description for 11fd0a6
  taronaeo

  authored
  11fd0a6View commit details

  Copy full SHA for 11fd0a6

  Browse repository at this point

• vulkan: Support Q2_0 (#25430)

  Show description for 788e07d
  jeffbolznv

  authored
  788e07dView commit details

  Copy full SHA for 788e07d

  Browse repository at this point

• sycl: fix row calculation when K_QUANTS_PER_ITERATION is 1 (#25690)

  Show description for 0bd0ec6
  malsbat

  authored
  0bd0ec6View commit details

  Copy full SHA for 0bd0ec6

  Browse repository at this point

• opencl: add ABS op (#25115)

  Gezahegne

  authored
  b85833eView commit details

  Copy full SHA for b85833e

  Browse repository at this point

Commits on Jul 16, 2026

• opencl: loads quants as uint for q4_K and q5_K flat mv (optimization for Adreno A7x GPUs) (#25780)

  Show description for e8f19cc
  

  wanghqc

  and

  lhez

  authored
  e8f19ccView commit details

  Copy full SHA for e8f19cc

  Browse repository at this point

• docs: added a note about using OpenCl with Adreno 810 (#25786)

  akleine

  authored
  ac2557cView commit details

  Copy full SHA for ac2557c

  Browse repository at this point

• DeepseekV4: Add fused hyper-connection ops (#25585)

  Show description for 0dc74e3
  am17an

  authored
  0dc74e3View commit details

  Copy full SHA for 0dc74e3

  Browse repository at this point

• hexagon: L2 cache handling rework (dirty bit tracking with lazy flushing) and more MUL_MAT updates (#25762)

  Show description for b2dd28a
  max-krasnyansky

  authored
  b2dd28aView commit details

  Copy full SHA for b2dd28a

  Browse repository at this point

• kleidiai: Add SME vs SME2 distinction in kernel dispatch (#25478)

  Show description for f15bd60
  matcraje

  authored
  f15bd60View commit details

  Copy full SHA for f15bd60

  Browse repository at this point

• vulkan: when using transfer queue for async copies, sync on event_wait to avoid race (#25229)

  0cc4m

  authored
  b15ca93View commit details

  Copy full SHA for b15ca93

  Browse repository at this point

• conversion: accept BitNetForCausalLM architecture name (#25769)

  Show description for 3278e92
  khashayarghafouri

  authored
  3278e92View commit details

  Copy full SHA for 3278e92

  Browse repository at this point

• TP: fix Phi3, Bert, Plamo2/3, ChatGLM (#25536)

  JohannesGaessler

  authored
  2e1fd76View commit details

  Copy full SHA for 2e1fd76

  Browse repository at this point

• vendor: update BoringSSL to 0.20260713.0 (#25624)

  cabelo

  authored
  86b719bView commit details

  Copy full SHA for 86b719b

  Browse repository at this point

• tests: actually exercise `test-recurrent-state-rollback` (#25758)

  am17an

  authored
  32e789fView commit details

  Copy full SHA for 32e789f

  Browse repository at this point

• server : allow text-only slot save/restore with mtmd (#25076)

  CHIPMUNK-T0T

  authored
  a8dc0e3View commit details

  Copy full SHA for a8dc0e3

  Browse repository at this point

• convert : fix dflash target tokenizer mismatch during conversion (#25733)

  Show description for a55a8c5
  ruixiang63

  authored
  a55a8c5View commit details

  Copy full SHA for a55a8c5

  Browse repository at this point

• CUDA: Support CUDA Virtual Devices (#25228)

  Show description for 79bba02
  anavp-nvidia

  authored
  79bba02View commit details

  Copy full SHA for 79bba02

  Browse repository at this point

• Enable CUDA graphs on volta+turing (#25749)

  heislera763

  authored
  3f08ef2View commit details

  Copy full SHA for 3f08ef2

  Browse repository at this point

• server: Ignore empty / non-existing `Origin` headers (#25756)

  Show description for 8ee54c8
  sdroege

  authored
  8ee54c8View commit details

  Copy full SHA for 8ee54c8

  Browse repository at this point

• ggml-cuda : restore prop.integrated on HIP builds (#24233)

  Show description for c7d8722
  liminfei-amd

  authored
  c7d8722View commit details

  Copy full SHA for c7d8722

  Browse repository at this point

• CUDA: dedup MoE gate/up activation quantization (#25441)

  Show description for 5839ba3
  praneshgo

  authored
  5839ba3View commit details

  Copy full SHA for 5839ba3

  Browse repository at this point

• ci : add official website link to release notes (#25728)

  Show description for a320cbf
  ggerganov

  authored
  a320cbfView commit details

  Copy full SHA for a320cbf

  Browse repository at this point

• quant : allow using manual tensor types with --pure (#25716)

  ggerganov

  authored
  56d6e9dView commit details

  Copy full SHA for 56d6e9d

  Browse repository at this point

• opencl: disable FA and MoE weights repack to work around compiler issues for Adreno 850 GPU (#25745)

  Show description for 3dafb58
  

  wanghqc

  and

  lhez

  authored
  3dafb58View commit details

  Copy full SHA for 3dafb58

  Browse repository at this point

• cuda: extract Q1_0 elements via __byte_perm (#25628)

  dfriehs

  authored
  602f828View commit details

  Copy full SHA for 602f828

  Browse repository at this point