All pull requests

Pull requests

• Bind trainer dest names at weight-sync initialize

  #278·sfc-gh-truwase opened Sep 19, 2026Collaborator·4 tasks done·

  ·

• Add Qwen3.5 HF-to-vLLM weight-sync adapters

  #277·sfc-gh-truwase opened Sep 16, 2026Collaborator·2 tasks done·

  ·
  1

• Pass max_spec_tokens to suffix_decode in the simulator

  #276·Dev-next-gen opened Sep 13, 2026·

  ·

• Fix AttributeError in SuffixDecodingCache.speculate when max_spec_tokens is None

  #275·Dev-next-gen opened Sep 13, 2026·

  ·

• [kernel generation] add kernel generation project to ArcticInference/projects/

  #273·sfc-gh-lpaille opened Sep 3, 2026·

  ·

• Tune rollout config: TP-aware FCA/SD thresholds

  #268·sfc-gh-mhidayetoglu opened Jun 29, 2026Contributor·

  ·
  4

• Multi-cache dynamic NTK RoPE

  #260·sfc-gh-yewang opened Apr 22, 2026Contributor·

  ·

• Ydu v14

  #251·sfc-gh-ydu opened Feb 24, 2026Contributor·

  ·
  1

• [Docs] Add vLLM version compatibility guide

  #235·yurekami opened Jan 2, 2026·3 tasks done·

  ·
  1

  2

• [Ulysses] Add validation for max_num_seqs divisibility by sp_size

  #234·yurekami opened Jan 2, 2026·2 tasks done·

  ·
  1

  1

• [Quality] Fix bare excepts, add CI workflow and contributor guidelines

  #233·yurekami opened Jan 2, 2026·2 of 3 tasks·

  ·

• [WIP]: Add the new kernels for optimizing moe inference

  #226·sfc-gh-reyazda opened Nov 25, 2025Contributor·

  ·

• update SP+EP

  #209·sfc-gh-mhidayetoglu opened Oct 15, 2025Contributor·

  ·
  14

• Support vllm 0.10.2

  #200·sfc-gh-goliaro opened Oct 2, 2025·

  ·
  2

• Add moe router module with hierarchical gating support for DeepSeek Models

  #191·sfc-gh-reyazda opened Sep 25, 2025Contributor·

  ·

• [WIP] SP+EP

  #181·sfc-gh-mhidayetoglu opened Sep 14, 2025Contributor·

  ·

• [gpt-oss] Fixes to work with vLLM v0.10.1

  #170·goliaro opened Sep 3, 2025·

  ·