Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit history

Commits on May 20, 2025

• model : fix llama4 graph (#13663)

  Show description for be02396
  ggerganov

  authored
  be02396

  Copy full SHA for be02396

• kv-cache : add SWA support (#13194)

  Show description for e298d2f
  ggerganov

  authored
  e298d2f

  Copy full SHA for e298d2f

Commits on May 14, 2025

• fix: Move build_inp_pos to the top of the graph section for build_granite (#13538)

  Show description for 5e7d95e
  gabe-l-hart

  authored
  5e7d95e

  Copy full SHA for 5e7d95e

Commits on May 13, 2025

• model : Granite MoE shared (#13269)

  Show description for d590cd4
  gabe-l-hart

  authored
  d590cd4

  Copy full SHA for d590cd4

Commits on May 12, 2025

• llama/ggml: add LLM training support (#10544)

  Show description for 10d2af0
  JohannesGaessler

  authored
  10d2af0

  Copy full SHA for 10d2af0

Commits on May 9, 2025

• llama : do not crash if there is no CPU backend (#13395)

  Show description for 27ebfca
  slaren

  authored
  27ebfca

  Copy full SHA for 27ebfca

• llama : one-off chat template fix for Mistral-Small-2503 (#13398)

  Show description for 3f96aef
  ngxson

  authored
  3f96aef

  Copy full SHA for 3f96aef

Commits on May 8, 2025

• context : allow cache-less context for embeddings (#13108)

  Show description for 6562e5a
  ggerganov

  authored
  6562e5a

  Copy full SHA for 6562e5a

• llama : print size and type of overridden tensors (#13364)

  slaren

  authored
  f061021

  Copy full SHA for f061021

Commits on May 7, 2025

• llama : deci : support ffn-free with attention (#13296)

  CISC

  authored
  bc4e112

  Copy full SHA for bc4e112

• llama : support tie embedding for chatglm models (#13328)

  piDack

  authored
  6c7fd67

  Copy full SHA for 6c7fd67

Commits on May 3, 2025

• llama : Llama-3_1-Nemotron-Ultra-253B-v1 support (#12843)

  ymcki

  authored
  3bf785f

  Copy full SHA for 3bf785f

Commits on May 2, 2025

• llama-model : support Qwen2 embedding models and pooling_mode_lasttoken (#13245)

  cebtenzzre

  authored
  2f56761

  Copy full SHA for 2f56761

• kv-cache : separate recurrent vs non-recurrent impl (#12799)

  Show description for c642bc0
  ggerganov

  authored
  c642bc0

  Copy full SHA for c642bc0

• llama : orion rope type is neox (#13261)

  CISC

  authored
  cb06a3c

  Copy full SHA for cb06a3c

• llama : plamo rope type is neox (#13260)

  CISC

  authored
  626083f

  Copy full SHA for 626083f

Commits on May 1, 2025

• llama-model : fix the reported size class for nomic-embed-text-v2-moe (#13223)

  cebtenzzre

  authored
  a70183e

  Copy full SHA for a70183e

Commits on Apr 29, 2025

• CUDA: fix non-cont. inputs for batched mat mul (#13155)

  JohannesGaessler

  authored
  cdf7658

  Copy full SHA for cdf7658

• llama : llm_type order by size (#13177)

  CISC

  authored
  7d3af70

  Copy full SHA for 7d3af70

• llama : set qwen3 model type sizes (#13175)

  CISC

  authored
  e98b369

  Copy full SHA for e98b369

Commits on Apr 28, 2025

• model : Nomic Embed Text V2 with Mixture-of-Experts (MoE) architecture (#12466)

  Show description for 5f5e39e
  

  manyoso

  and

  cebtenzzre

  authored
  5f5e39e

  Copy full SHA for 5f5e39e

• CUDA: fix q_nope_absorbed prec for DS 2 Lite f16 (#13137)

  JohannesGaessler

  authored
  69699be

  Copy full SHA for 69699be

Commits on Apr 17, 2025

• graph : make FA compatible with MLA + add initial Metal kernels (#12953)

  Show description for 2f74c35
  ggerganov

  authored
  2f74c35

  Copy full SHA for 2f74c35

Commits on Apr 15, 2025

• llama : DeepSeek V2/V3 MLA implementation (#12801)

  Show description for daa4228
  jukofyork

  authored
  daa4228

  Copy full SHA for daa4228

Commits on Apr 11, 2025

• llama-model : add Glm4Model implementation for GLM-4-0414 (#12867)

  Show description for 06bb53a
  zRzRzRzRzRzRzR

  authored
  06bb53a

  Copy full SHA for 06bb53a

• llama : correct rms norm for llama 4 (#12882)

  ngxson

  authored
  8b91d53

  Copy full SHA for 8b91d53

Commits on Apr 9, 2025

• llama : Support Qwen3 and Qwen3MoE (#12828)

  Show description for d3bd719
  

  bozheng-hit

  and

  bozheng-hit

  authored
  d3bd719

  Copy full SHA for d3bd719

Commits on Apr 7, 2025

• llama : Support llama 4 text-only (#12791)

  Show description for 1466621
  ngxson

  authored
  1466621

  Copy full SHA for 1466621

Commits on Apr 2, 2025

• llama : add option to override model tensor buffers (#11397)

  Show description for e0e912f
  slaren

  authored
  e0e912f

  Copy full SHA for e0e912f

Commits on Mar 30, 2025

• llama : support BailingMoE (Ling) (#12634)

  CISC

  authored
  2c3f8b8

  Copy full SHA for 2c3f8b8

Commits on Mar 29, 2025

• llama : change cpu_buft_list order: ACCEL -> GPU host -> CPU extra -> CPU (#12632)

  Show description for 0bb2919
  

  Djip007

  and

  philou

  authored
  0bb2919

  Copy full SHA for 0bb2919

Commits on Mar 28, 2025

• llama : fix incorrect Qwen2Moe ffn_moe_out graph callback (#12631)

  CISC

  authored
  3714c3e

  Copy full SHA for 3714c3e

Commits on Mar 27, 2025

• llama : add PLM GGUF Conversion & Inference Support (#12457)

  Show description for f125b8d
  Si1w

  authored
  f125b8d

  Copy full SHA for f125b8d

• model : restore support for T5Encoder (#12590)

  HighDoping

  authored
  953c2a6

  Copy full SHA for 953c2a6

Commits on Mar 22, 2025

• llama : gemma3 : use output tensor if it exists in model weight (#12506)

  Show description for fbdfefe
  ngxson

  authored
  fbdfefe

  Copy full SHA for fbdfefe