Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 7, 2025

• memory : use sequential equal splits for recurrent modules (#16442)

  ggerganov

  authored
  0123ff3

  Copy full SHA for 0123ff3

• metal : add support for non-padded FA KV (#16148)

  Show description for 0a319bb
  ggerganov

  authored
  0a319bb

  Copy full SHA for 0a319bb

• tests : add -INF blocks to the KQ mask in the FA tests (#16380)

  Show description for 1d6092f
  

  ggerganov

  and

  jeffbolznv

  authored
  1d6092f

  Copy full SHA for 1d6092f

• metal : various optimizations + refactoring (#16446)

  Show description for 8ae32dc
  ggerganov

  authored
  8ae32dc

  Copy full SHA for 8ae32dc

Commits on Oct 6, 2025

• llama : add --no-host to disable host buffers (#16310)

  Show description for 3df2244
  

  Gadflyii

  and

  slaren

  authored
  3df2244

  Copy full SHA for 3df2244

• chat : Granite Docling stopping (#16438)

  Show description for c08002a
  gabe-l-hart

  authored
  c08002a

  Copy full SHA for c08002a

• ci : refactor sdk caching to minimize storage (#16414)

  Show description for 3a002af
  CISC

  authored
  3a002af

  Copy full SHA for 3a002af

• ggml : fix unaligned access in AMX code (#16315)

  ggerganov

  authored
  a23b9bd

  Copy full SHA for a23b9bd

• ci : remove missing reranker model files (#16444)

  Show description for 04e632a
  danbev

  authored
  04e632a

  Copy full SHA for 04e632a

• ggml-cpu : fix leftover handling in ggml_vec_scale_f32 for SVE (#16443)

  Show description for a80ff18
  danbev

  authored
  a80ff18

  Copy full SHA for a80ff18

• nix : removed metal for nix (#16118)

  yuannan

  authored
  1d49ca3

  Copy full SHA for 1d49ca3

• server: update readme to mention n_past_max metric (#16436)

  Show description for c5fef0f
  okuvshynov

  authored
  c5fef0f

  Copy full SHA for c5fef0f

Commits on Oct 5, 2025

• model : Granite docling + Idefics3 preprocessing (SmolVLM) (#16206)

  Show description for ca71fb9
  

  gabe-l-hart

  and

  ngxson

  authored
  ca71fb9

  Copy full SHA for ca71fb9

• ggml webgpu: actually add softmax, fix rms_norm offset (#16400)

  Show description for 3526657
  reeselevine

  authored
  3526657

  Copy full SHA for 3526657

Commits on Oct 4, 2025

• vulkan: use a more appropriate amount of threads when generating shaders (#16418)

  Show description for 86df2c9
  netrunnereve

  authored
  86df2c9

  Copy full SHA for 86df2c9

• rpc : check src buffer when copying tensor (#16421)

  Show description for f392839
  rgerganov

  authored
  f392839

  Copy full SHA for f392839

• rpc : add support for multiple devices (#16276)

  Show description for 898acba
  rgerganov

  authored
  898acba

  Copy full SHA for 898acba

• vulkan : incremental shader builds (#16341)

  Show description for e29acf7
  

  Acly

  and

  jeffbolznv

  authored
  e29acf7

  Copy full SHA for e29acf7

Commits on Oct 3, 2025

• chat : support Magistral thinking (#16413)

  Show description for 128d522
  ServeurpersoCom

  authored
  128d522

  Copy full SHA for 128d522

• server : context checkpointing for hybrid and recurrent models (#16382)

  Show description for f6dcda3
  

  ddh0

  and

  ggerganov

  authored
  f6dcda3

  Copy full SHA for f6dcda3

• metal : fix loop bound in ggml_mem_ranges (#16412)

  ggerganov

  authored
  606a73f

  Copy full SHA for 606a73f

• llama : fix shapes for bert/mpt q/k norm (#16409)

  CISC

  authored
  946f71e

  Copy full SHA for 946f71e

• ggml : fix graph reallocation with multiple chunks (#16396)

  Show description for 638d330
  Acly

  authored
  638d330

  Copy full SHA for 638d330

• Fix missing messages on sibling navigation (#16408)

  Show description for 84c8e30
  allozaur

  authored
  84c8e30

  Copy full SHA for 84c8e30

• vulkan: Replace uses of maxMemoryAllocationSize and VK_WHOLE_SIZE (#16354)

  Show description for 2aaf0a2
  jeffbolznv

  authored
  2aaf0a2

  Copy full SHA for 2aaf0a2

• vulkan: Fix FA coopmat1 invalid array indexing (#16365)

  Show description for 0e1f838
  jeffbolznv

  authored
  0e1f838

  Copy full SHA for 0e1f838

• ci : change macos-13 to macos-15-intel (#16401)

  Show description for ad12647
  danbev

  authored
  ad12647

  Copy full SHA for ad12647

• Capture model name only after first token (streaming) or completed request (#16405)

  Show description for 7723327
  allozaur

  authored
  7723327

  Copy full SHA for 7723327

• vulkan: in flash attention, bounds check against nem1 (don't rely on GGML_KQ_MASK_PAD) (#16316)

  jeffbolznv

  authored
  e308efd

  Copy full SHA for e308efd

• webui : Fix messages payload sent to chat completions (#16402)

  Show description for 136bda7
  allozaur

  authored
  136bda7

  Copy full SHA for 136bda7

• fix: track viewportHeight via window.innerHeight to avoid unwanted scrolling (#16356)

  Show description for 5113efd
  

  ServeurpersoCom

  and

  allozaur

  authored
  5113efd

  Copy full SHA for 5113efd

Commits on Oct 2, 2025

• test-barrier : do not use more threads than physically available (#16389)

  Show description for d64c810
  

  CISC

  and

  jeffbolznv

  authored
  d64c810

  Copy full SHA for d64c810

• ggml webgpu: add support for soft_max, optimize rms_norm (#16357)

  Show description for ef07a40
  

  reeselevine

  and

  ggerganov

  authored
  ef07a40

  Copy full SHA for ef07a40

• model : Apertus model implementation (#15852)

  Show description for 34fcc5a
  

  4 peopleauthored
  34fcc5a

  Copy full SHA for 34fcc5a

• musa: update compile flags (#16265)

  Show description for 91a2a56
  yeahdongcn

  authored
  91a2a56

  Copy full SHA for 91a2a56

Pagination

Previous

Next