Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on May 30, 2026

• TP: fix granularity for Qwen 3.5/3.6 + 3 GPUs (#23843)

  Show description for 8b0e0db
  JohannesGaessler

  authored
  8b0e0db

  Copy full SHA for 8b0e0db

• llama : do not skip iGPU when only RPC devices are present (#23868)

  Show description for 1738129
  rgerganov

  authored
  1738129

  Copy full SHA for 1738129

Commits on May 29, 2026

• graph : ensure DS32 kq_mask_lid is F32 (#23864)

  CISC

  authored
  764f1e6

  Copy full SHA for 764f1e6

• model : support for DeepseekV32ForCausalLM with generic DeepSeek Sparse Attention (DSA) implementation (#23346)

  Show description for 1f0aa2a
  

  4 peopleauthored
  1f0aa2a

  Copy full SHA for 1f0aa2a

• llama: use f16 mask for FA to save VRAM (#23764)

  Show description for 031ddb2
  am17an

  authored
  031ddb2

  Copy full SHA for 031ddb2

• llama: add llm_graph_input_mtp (#23643)

  Show description for eef59a7
  

  am17an

  and

  ggerganov

  authored
  eef59a7

  Copy full SHA for eef59a7

Commits on May 28, 2026

• chat : add Granite 4.1 chat template (#23518)

  jesus-talavera-ibm

  authored
  bb771cb

  Copy full SHA for bb771cb

Commits on May 27, 2026

• convert: add MiniCPM5 tokenizer support (#23384)

  Show description for 9777256
  

  zhangtao2-1

  and

  zhangtao

  authored
  9777256

  Copy full SHA for 9777256

Commits on May 26, 2026

• models : Attach Mistral3 NVFP4 weight scales (#23629)

  michaelw9999

  authored
  6fe90de

  Copy full SHA for 6fe90de

• model : add support for talkie-1930-13b (#22596)

  Show description for c9d9829
  niklassheth

  authored
  c9d9829

  Copy full SHA for c9d9829

Commits on May 25, 2026

• model: tag ffn_latent as MUL_MAT to fix buft probe (#23664)

  Show description for 328874d
  ServeurpersoCom

  authored
  328874d

  Copy full SHA for 328874d

Commits on May 23, 2026

• model : add NVFP4 MTP scale tensors (#23563)

  Show description for b0df4c0
  michaelw9999

  authored
  b0df4c0

  Copy full SHA for b0df4c0

Commits on May 22, 2026

• vocab : fix HybridDNA tokenizer (#23466)

  Show description for afcda09
  

  kashif

  and

  CISC

  authored
  afcda09

  Copy full SHA for afcda09

Commits on May 21, 2026

• mtp: use inp_out_ids for skipping logit computation (#23433)

  Show description for 12e5d99
  am17an

  authored
  12e5d99

  Copy full SHA for 12e5d99

• vocab : add Carbon-3B (HybridDNATokenizer) support (#23410)

  Show description for 7ea23dd
  

  kashif

  and

  CISC

  authored
  7ea23dd

  Copy full SHA for 7ea23dd

• llama-graph: fix null-buffer crash in llm_graph_input_attn_kv_iswa for SWA-only models (#23131)

  Show description for eeeaf61
  ssfdre38

  authored
  eeeaf61

  Copy full SHA for eeeaf61

Commits on May 20, 2026

• mtmd, model : merge HunyuanOCR into HunyuanVL and fix OCR vision precision (#23329)

  Show description for 6a257d4
  wendadawen

  authored
  6a257d4

  Copy full SHA for 6a257d4

• Move to backend sampling for MTP draft path (#23287)

  Show description for ad27757
  gaugarg-nv

  authored
  ad27757

  Copy full SHA for ad27757

• metal : optimize pad + cpy (#23354)

  Show description for 57ebaf4
  ggerganov

  authored
  57ebaf4

  Copy full SHA for 57ebaf4

Commits on May 19, 2026

• model : clarify MTP layer comment in qwen35.cpp [no ci] (#23338)

  Show description for baf3cc6
  danbev

  authored
  baf3cc6

  Copy full SHA for baf3cc6

• llama : MTP clean-up (#23269)

  Show description for d14ce3d
  ggerganov

  authored
  d14ce3d

  Copy full SHA for d14ce3d

Commits on May 18, 2026

• llama: initialize pre-norm embedding mask flag (#23256)

  abetlen

  authored
  49c21f9

  Copy full SHA for 49c21f9

Commits on May 17, 2026

• llama: avoid copying logits during prompt decode in MTP (#23198)

  Show description for 3e12fbd
  am17an

  authored
  3e12fbd

  Copy full SHA for 3e12fbd

Commits on May 16, 2026

• llama + spec: MTP Support (#22673)

  Show description for 2555826
  

  am17an

  and

  ggerganov

  authored
  2555826

  Copy full SHA for 2555826

• model : NvFP4 quantized LM head support (#23046)

  Show description for 42928bc
  ynankani

  authored
  42928bc

  Copy full SHA for 42928bc

Commits on May 14, 2026

• unicode,test: add Qwen3.5 non-backtracking tokenizer handler and regr… (#22110)

  Show description for 42532af
  

  3 peopleauthored
  42532af

  Copy full SHA for 42532af

Commits on May 11, 2026

• spec : parallel drafting support (#22838)

  Show description for 68e7ea3
  

  ggerganov

  and

  Petros Sideris

  authored
  68e7ea3

  Copy full SHA for 68e7ea3

Commits on May 10, 2026

• model : fix model type check for granite/llama3 and deepseek2/glm4.7 lite (#22870)

  CISC

  authored
  5755a10

  Copy full SHA for 5755a10

Commits on May 9, 2026

• model : add sarvam_moe architecture support (#20275)

  sumitchatterjee13

  authored
  1e5ad35

  Copy full SHA for 1e5ad35

Commits on May 8, 2026

• model : support Gemma4_26B_A4B_NVFP4 (#22804)

  Show description for 9f5f0e6
  

  ynankani

  and

  CISC

  authored
  9f5f0e6

  Copy full SHA for 9f5f0e6

Commits on May 7, 2026

• llama : fix device state save/load (#22805)

  ggerganov

  authored
  e43431b

  Copy full SHA for e43431b

• llama : remove unnecessary seq_id check during state restore (#22797)

  ggerganov

  authored
  803627f

  Copy full SHA for 803627f

• model: Add Mimo v2.5 model support (#22493)

  Show description for 8e52631
  

  AesSedai

  and

  CISC

  authored
  8e52631

  Copy full SHA for 8e52631

• llama : add missing call to ggml_backend_load_all() (#22752)

  Show description for 3980e04
  angt

  authored
  3980e04

  Copy full SHA for 3980e04

Commits on May 6, 2026

• model : don't crash on unsupported architecture (#22742)

  Show description for 5207d12
  

  giladgd

  and

  CISC

  authored
  5207d12

  Copy full SHA for 5207d12