All pull requests

Pull requests

• chore: CI proxy for #4663

  #4703·akoumpa opened Nov 9, 2026Contributor·

  ·
  3

• chore: CI proxy for #4552

  #4702·akoumpa opened Nov 9, 2026Contributor·

  ·
  2

• chore: CI proxy for #4652

  #4699·akoumpa opened Nov 9, 2026Contributor·

  ·
  2

• fix(kimi_linear): apply expert renames to PEFT adapter keys and targets

  community-request
  #4697·0z5a opened Nov 9, 2026Contributor·2 of 3 tasks·

  ·
  1

  2

• fix(peft): compute DoRA magnitude from loaded base weights

  community-request
  #4696·0z5a opened Nov 9, 2026Contributor·2 of 3 tasks·

  ·
  1

  2

• fix(docs): support retrieval family and guide routing

  #4693·oliverholworthy opened Nov 9, 2026Contributor·

  ·
  1

• feat(dflash): add DSpine draft training

  community-request
  #4692·cloudnathan5 opened Nov 9, 2026·3 tasks done·

  ·
  1

• fix(dflash): honor draft config overrides for Qwen3 targets

  community-request
  #4691·cloudnathan5 opened Nov 9, 2026·3 tasks done·

  ·
  2

• refactor(audio): type normalized waveforms and ASR processor requests

  community-request
  #4690·Mist-wu opened Nov 9, 2026·

  ·
  2

• refactor(optim): type initial parameter groups and override records

  community-request
  #4689·Mist-wu opened Nov 9, 2026·

  ·
  2

• refactor(eval): type manual decoding and prompt boundaries

  community-request
  #4688·Mist-wu opened Nov 9, 2026·

  ·
  2

• refactor(speculative): type benchmark request and JSON contracts

  community-request
  #4687·Mist-wu opened Nov 9, 2026·3 tasks done·

  ·
  2

• feat(speculative): add gated-LoRA MTP

  community-request
  #4686·kashif opened Nov 9, 2026Contributor·3 tasks done·

  ·
  4

• fix(minimax_m3_vl): check the released DSpark vocab_size against the target

  community-request
  #4685·ischencheng opened Nov 9, 2026Contributor·2 of 3 tasks·

  ·
  1

• fix(datasets): raise when NanogptDataset can never yield a sample

  community-request
  #4684·pcbeingused333 opened Nov 9, 2026Contributor·2 of 3 tasks·

  ·
  1

• fix(nemotron): preserve media MTP, LoRA gradients, and packed FSDP

  #4683·HuiyingLi opened Nov 8, 2026Contributor·5 tasks done·

  ·
  1

• fix(vlm): account for PP and TP in per-GPU throughput

  #4682·HuiyingLi opened Nov 8, 2026Contributor·4 tasks done·

  ·
  2

• perf(gpt-oss): support packed FA4 training with learned sinks

  #4681·akoumpa opened Nov 8, 2026Contributor·5 tasks done·

  ·
  3

• feat(deepseek-v41): add configurable pipeline parallelism

  #4680·HuiyingLi opened Nov 8, 2026Contributor·3 of 4 tasks·

  ·
  3

• refactor(moe): type load-balance metric records

  community-request
  #4677·shu-chen-dot-ca opened Nov 7, 2026·2 of 3 tasks·

  ·
  2

• fix(checkpoint): accept SerializationFormat in model_save_format

  community-request
  #4676·edjson opened Nov 7, 2026Contributor·2 of 3 tasks·

  ·
  1

• chore(types): tighten cuda_graphs partial capture_owner and target types

  community-request
  #4675·edjson opened Nov 7, 2026Contributor·3 tasks done·

  ·
  1

• perf(qwen3): compile projections around packed FA4 attention

  #4670·akoumpa opened Nov 7, 2026Contributor·5 of 7 tasks·

  ·
  4

• fix(config): require NEMO_ENABLE_USER_MODULES for file.py targets

  #4668·akoumpa opened Nov 7, 2026Contributor·3 tasks done·

  ·
  3

• ci: proxy CI run for PR #4657

  #4664·HuiyingLi opened Nov 6, 2026Contributor·

  ·
  4