Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Jan 26, 2026

• graph : fix nkvo offload with FA (#19105)

  ggerganov

  authored
  8f80d1b

  Copy full SHA for 8f80d1b

Commits on Jan 25, 2026

• kv-cache : support V-less cache (#19067)

  Show description for d9c6ce4
  ggerganov

  authored
  d9c6ce4

  Copy full SHA for d9c6ce4

Commits on Jan 23, 2026

• graph : utilize `ggml_build_forward_select()` to avoid reallocations (#18898)

  Show description for 557515b
  ggerganov

  authored
  557515b

  Copy full SHA for 557515b

Commits on Jan 22, 2026

• mla : make the V tensor a view of K (#18986)

  Show description for a5eaa1d
  

  ggerganov

  and

  JohannesGaessler

  authored
  a5eaa1d

  Copy full SHA for a5eaa1d

Commits on Jan 21, 2026

• memory : add llama_memory_hybrid_iswa (#18601)

  Show description for ad8d85b
  

  tdakhran

  and

  ggerganov

  authored
  ad8d85b

  Copy full SHA for ad8d85b

Commits on Jan 13, 2026

• graph : clean up t5 input builders (#18795)

  Show description for 076b0fa
  gabe-l-hart

  authored
  076b0fa

  Copy full SHA for 076b0fa

Commits on Jan 5, 2026

• model : add LFM2-ColBert-350M (#18607)

  Show description for 73d284a
  tdakhran

  authored
  73d284a

  Copy full SHA for 73d284a

• models : fix backend assignment for Granite/Nemotron graphs (#18599)

  Show description for 2da64a2
  ggerganov

  authored
  2da64a2

  Copy full SHA for 2da64a2

Commits on Jan 4, 2026

• sampling : add support for backend sampling (#17004)

  Show description for d3dce4e
  

  4 peopleauthored
  d3dce4e

  Copy full SHA for d3dce4e

Commits on Jan 3, 2026

• graph : fix graph reuse logic when `n_pos_per_embd > 1` (#18566)

  ggerganov

  authored
  c69c7eb

  Copy full SHA for c69c7eb

Commits on Dec 16, 2025

• graph : reuse SSM graphs (#16490)

  Show description for c560316
  ggerganov

  authored
  c560316

  Copy full SHA for c560316

• llama : add support for NVIDIA Nemotron 3 Nano (#18058)

  Show description for 2995341
  

  danbev

  and

  ggerganov

  authored
  2995341

  Copy full SHA for 2995341

Commits on Dec 14, 2025

• graph: add f_attn_temp_offset (#18025)

  

  Xuan-Son Nguyen

  authored
  0759b09

  Copy full SHA for 0759b09

• models : fix YaRN regression + consolidate logic (#18006)

  Show description for 609a2d0
  ggerganov

  authored
  609a2d0

  Copy full SHA for 609a2d0

Commits on Dec 12, 2025

• models : fix the attn_factor for mistral3 graphs + improve consistency (#17945)

  Show description for 7bed317
  ggerganov

  authored
  7bed317

  Copy full SHA for 7bed317

Commits on Dec 10, 2025

• ggml : remove GGML_KQ_MASK_PAD constant (#17910)

  Show description for 4dff236
  ggerganov

  authored
  4dff236

  Copy full SHA for 4dff236

Commits on Dec 8, 2025

• graph : use fill instead of scale_bias in grouped expert selection (#17867)

  Show description for c8554b6
  CISC

  authored
  c8554b6

  Copy full SHA for c8554b6

Commits on Dec 1, 2025

• model: support Ministral3 (#17644)

  Show description for cd3c118
  

  3 peopleauthored
  cd3c118

  Copy full SHA for cd3c118

• llama-graph: avoid expand_forward for fusion (#17633)

  am17an

  authored
  6eea666

  Copy full SHA for 6eea666

Commits on Nov 25, 2025

• ggml : add ggml_top_k (#17365)

  Show description for 583cb83
  ggerganov

  authored
  583cb83

  Copy full SHA for 583cb83

Commits on Nov 13, 2025

• CUDA: fuse rope + set_rows (#16884)

  Show description for a90eb94
  am17an

  authored
  a90eb94

  Copy full SHA for a90eb94

Commits on Nov 7, 2025

• hparams : add n_embd_inp() to support extended embed (#16928)

  Show description for 9008027
  CISC

  authored
  9008027

  Copy full SHA for 9008027

Commits on Oct 30, 2025

• llama : use std::abs instead of abs (#16853)

  kaetemi

  authored
  d739511

  Copy full SHA for d739511

Commits on Oct 26, 2025

• graph : add clamping to ffn_moe_weights_sum to avoid div-by-zero (#16655)

  Show description for f696428
  CISC

  authored
  f696428

  Copy full SHA for f696428

• CUDA: General GEMV fusion (#16715)

  am17an

  authored
  f77c13b

  Copy full SHA for f77c13b

Commits on Oct 20, 2025

• model : add BailingMoeV2 support (#16063)

  Show description for 84bf3c6
  CISC

  authored
  84bf3c6

  Copy full SHA for 84bf3c6

Commits on Oct 13, 2025

• metal : FA support F32 K and V and head size = 32 (#16531)

  Show description for e60f241
  ggerganov

  authored
  e60f241

  Copy full SHA for e60f241

• graph : support cacheless embeddings with FA and iSWA (#16528)

  Show description for e38b7c6
  ggerganov

  authored
  e38b7c6

  Copy full SHA for e38b7c6

Commits on Oct 9, 2025

• model: EmbeddingGemma Adding Support for SentenceTransformers Dense Modules (#16367)

  Show description for e08db42
  

  3 peopleauthored
  e08db42

  Copy full SHA for e08db42

Commits on Sep 25, 2025

• model : add GroveMoE support (#15510)

  Show description for 835b2b9
  CISC

  authored
  835b2b9

  Copy full SHA for 835b2b9

• CUDA: add a fused top-K MoE kernel (#16130)

  Show description for 077c94d
  am17an

  authored
  077c94d

  Copy full SHA for 077c94d

• llama : add support for qwen3 reranker (#15824)

  iamlemec

  authored
  b5bd037

  Copy full SHA for b5bd037

Commits on Sep 14, 2025

• model : add grok-2 support (#15539)

  Show description for b8e09f0
  CISC

  authored
  b8e09f0

  Copy full SHA for b8e09f0

Commits on Sep 10, 2025

• graph : support non-contiguous Q in build_attn_mha (#15908)

  Show description for 6ab397e
  

  CISC

  and

  ggerganov

  authored
  6ab397e

  Copy full SHA for 6ab397e

Commits on Sep 8, 2025

• context : fix n_outputs during reserve (#15858)

  Show description for 663027f
  ggerganov

  authored
  663027f

  Copy full SHA for 663027f