All pull requests

Pull requests

• Add EmbeddingGemma 2 full support

  #684·aleroot opened Oct 7, 2026Contributor·5 tasks done·

  ·

• fix loratrain for tiny prompts

  needs-changes
  #683·davidkoski opened Oct 7, 2026Member·5 tasks done·

  ·

• scripted Tokenizer

  cat-miscneeds-review
  #680·davidkoski opened Oct 5, 2026Member·5 tasks done·

  ·
  3

• Implement reranker protocols(Qwen variations)

  #676·aleroot opened Oct 4, 2026Contributor·4 of 5 tasks·

  ·
  1

  1

• Reuse Gemma4 softcap fusion for text inference

  #675·aleroot opened Oct 4, 2026Contributor·3 of 5 tasks·

  ·

• Automatic prefill optimization

  cat-kv-cacheneeds-review
  #674·aleroot opened Oct 4, 2026Contributor·4 of 5 tasks·

  ·

• Add Qwen3-VL text and image embeddings

  cat-modelneeds-review
  #665·aleroot opened Oct 2, 2026Contributor·4 of 5 tasks·

  ·

• Add compiled decode for Qwen3.5 VLM models

  cat-modelneeds-review
  #662·ngutech21 opened Oct 1, 2026Contributor·5 tasks done·

  ·
  3

• Resume a hybrid model's cache across turns from a checkpoint

  cat-kv-cacheneeds-review
  #661·jhancock1975 opened Oct 1, 2026Contributor·5 tasks done·

  ·
  1

• Parallelize gated-delta training chunks

  cat-modelneeds-review
  #656·aleroot opened Sep 30, 2026Contributor·4 of 5 tasks·

  ·

• Report the reused cache prefix before prefill

  cat-kv-cacheneeds-review
  #644·aleroot opened Sep 27, 2026Contributor·5 tasks done·

  ·

• Add bounded prompt lookup decoding

  cat-samplingneeds-review
  #643·aleroot opened Sep 27, 2026Contributor·5 tasks done·

  ·

• Add MiniCPM5 Tool Call Parser

  cat-toolsneeds-review
  #637·ibrahimcetin opened Sep 24, 2026·5 tasks done·

  ·

• Add LFM 2.5 2.6b Support

  cat-modelneeds-review
  #634·ibrahimcetin opened Sep 23, 2026·5 tasks done·

  ·
  1

• Load only the shard the safetensors index maps a key prefix to

  cat-samplingneeds-review
  #632·spokvulcan opened Sep 20, 2026Contributor·4 of 5 tasks·

  ·
  1

• feat(models): add Bonsai 2 27B support

  cat-modelneeds-review
  #630·spokvulcan opened Sep 19, 2026Contributor·4 of 5 tasks·

  ·
  1

• Fix exact prefix rewinds in rotating KV caches

  cat-kv-cacheneeds-review
  #622·aleroot opened Sep 14, 2026Contributor·5 tasks done·

  ·
  2

  

• Allow tokenizers to provide streaming detokenizers

  cat-toolsneeds-review
  #619·aleroot opened Sep 11, 2026Contributor·4 of 5 tasks·

  ·
  7

• Add mid-turn steering to ChatSession

  cat-miscneeds-review
  #608·aleroot opened Sep 5, 2026Contributor·5 tasks done·

  ·

• DFlash2 speculative decoding for Qwen3.8-27B

  cat-samplingneeds-review
  #607·spokvulcan opened Sep 4, 2026Contributor·5 tasks done·

  ·
  1

  1

• Expose prepared categorical distributions for samplers

  cat-samplingneeds-review
  #595·VaderChen opened Aug 30, 2026·5 tasks done·

  ·
  6

• Add resumable Qwen MTP speculative decoding to ChatSession

  cat-samplingneeds-ci
  #581·ngutech21 opened Aug 26, 2026Contributor·4 tasks done·

  ·
  11

• Load model components on demand, starting with MuseGlimmer's vision encoder

  cat-modelneeds-review
  #580·aleroot opened Aug 26, 2026Contributor·3 of 4 tasks·

  ·
  11

  

• Add reasoning streaming events and support Gemma 4 channels

  cat-toolsneeds-review
  #563·abra-code opened Aug 21, 2026Contributor·4 tasks done·

  ·
  8

  

• Report the progress of reading the weights while loading a model

  cat-miscneeds-changes
  #551·aleroot opened Aug 18, 2026Contributor·4 tasks done·

  ·
  1