Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 1, 2025

• Mark some multi-prediction tensors as not required. (#814)

  Downtown-Case

  authored
  6051ba2

  Copy full SHA for 6051ba2

Commits on Sep 30, 2025

• Attempt to fix AVX2 FA (#807)

  Show description for e94d1a9
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  e94d1a9

  Copy full SHA for e94d1a9

Commits on Sep 27, 2025

• Fix gemma3 vision (#803)

  Show description for 3d4977c
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  3d4977c

  Copy full SHA for 3d4977c

• Move minja and nlohmann/json to vendor (#802)

  Show description for 95780cd
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  95780cd

  Copy full SHA for 95780cd

• Remove stb_image.h copy in common - it is now in vendor (#801)

  Show description for 5064ff8
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  5064ff8

  Copy full SHA for 5064ff8

• Port mdmd from mainline + Qwen2/2.5-VL support (#798)

  Show description for 87e4762
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  87e4762

  Copy full SHA for 87e4762

Commits on Sep 26, 2025

• sync: vendor (#799)

  Show description for 367654f
  firecoperana

  authored
  367654f

  Copy full SHA for 367654f

• CPU: faster FA (#797)

  Show description for c108e4b
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  c108e4b

  Copy full SHA for c108e4b

Commits on Sep 25, 2025

• Fix logprobs (#787)

  Show description for 6bb76b1
  

  sayap

  and

  Iwan Kawrakow

  authored
  6bb76b1

  Copy full SHA for 6bb76b1

Commits on Sep 24, 2025

• Fused matrix multiplications (CUDA and CPU) (#796)

  Show description for 8e497e7
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  8e497e7

  Copy full SHA for 8e497e7

• Fix dequantization when requantizing (#795)

  Show description for 0d1bbde
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  0d1bbde

  Copy full SHA for 0d1bbde

• cpu: fused softmax+topk (#794)

  Show description for cde2eb5
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  cde2eb5

  Copy full SHA for cde2eb5

• Update webui to handle reasoning content and include usage stats in server only when requested (#791)

  Show description for 09db3a4
  firecoperana

  authored
  09db3a4

  Copy full SHA for 09db3a4

Commits on Sep 23, 2025

• Fix #772 (#790)

  Show description for 45afaf3
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  45afaf3

  Copy full SHA for 45afaf3

• model : add grok-2 support (#782)

  Show description for 8cd2d7c
  firecoperana

  authored
  8cd2d7c

  Copy full SHA for 8cd2d7c

• cuda: fused top_k+softmax as used in most MoE models (#789)

  Show description for 18f0435
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  18f0435

  Copy full SHA for 18f0435

• Fix compiler warnings (#788)

  Show description for 64e357c
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  64e357c

  Copy full SHA for 64e357c

Commits on Sep 13, 2025

• Deepseek V3.1 native tool calling support (OpenAI Style) (#771)

  firecoperana

  authored
  6d2e7ca

  Copy full SHA for 6d2e7ca

Commits on Sep 11, 2025

• fix convert error for ernie 4.5 (#774)

  firecoperana

  authored
  0ce2f93

  Copy full SHA for 0ce2f93

Commits on Sep 9, 2025

• fix v1 completions streaming mode (#768)

  firecoperana

  authored
  d323871

  Copy full SHA for d323871

Commits on Sep 5, 2025

• This is very slightly better (#762)

  Show description for c519d41
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  c519d41

  Copy full SHA for c519d41

• Fix ggml_is_contiguously_allocated (#764)

  Show description for c15f8ac
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  c15f8ac

  Copy full SHA for c15f8ac

• Add Ernie 4.5 MOE and 0.3B Support (#759)

  Show description for 33e0712
  firecoperana

  authored
  33e0712

  Copy full SHA for 33e0712

• llama: enable K-shift for quantized KV cache for cuda (#760)

  Show description for cec8b70
  

  firecoperana

  and

  Nekotekina

  authored
  cec8b70

  Copy full SHA for cec8b70

Commits on Sep 4, 2025

• Offload only activated experts to the GPU (#698)

  Show description for 0c15494
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  0c15494

  Copy full SHA for 0c15494

• Better CPU SWA (#757)

  Show description for 06cc7c6
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  06cc7c6

  Copy full SHA for 06cc7c6

• Alternative CUDA FA for SWA models (#754)

  Show description for f5e68bf
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  f5e68bf

  Copy full SHA for f5e68bf

Commits on Sep 2, 2025

• Refactor CUDA flash attention (#745)

  Show description for 3433c7b
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  3433c7b

  Copy full SHA for 3433c7b

• Set default value of GGML_SCHED_MAX_COPIES to 1 (#751)

  Show description for 1f43463
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  1f43463

  Copy full SHA for 1f43463

• Revert "CUDA: prompt processing optimizations for MoE models (#739)" (#748)

  Show description for 62f5382
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  62f5382

  Copy full SHA for 62f5382

Commits on Sep 1, 2025

• Remove double definition of LLAMA_LOG_DEBUG

  

  Iwan Kawrakow

  committed
  d10d90a

  Copy full SHA for d10d90a

• Tool calls support from mainline (#723)

  Show description for 0f9ecae
  

  firecoperana

  and

  Iwan Kawrakow

  authored
  0f9ecae

  Copy full SHA for 0f9ecae

Commits on Aug 31, 2025

• Fused FFN_UP+FFN_GATE op (#741)

  Show description for b66cecc
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  b66cecc

  Copy full SHA for b66cecc

Commits on Aug 30, 2025

• CUDA: prompt processing optimizations for MoE models (#739)

  Show description for f22a9ef
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  f22a9ef

  Copy full SHA for f22a9ef

Commits on Aug 29, 2025

• Sanitize imatrix (#735)

  Show description for 46968d4
  

  ikawrakow

  and

  Iwan Kawrakow

  authored
  46968d4

  Copy full SHA for 46968d4

Pagination

Previous

Next