All pull requests

Pull requests

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

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

  ·
  3

• perf(attention): reuse TE plans for packed sequences

  #4669·akoumpa opened Nov 7, 2026Contributor·6 of 8 tasks·

  ·
  3

• 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 #4499

  #4667·HuiyingLi opened Nov 6, 2026Contributor·

  ·
  3

• ci: proxy CI run for PR #4634

  #4665·HuiyingLi opened Nov 6, 2026Contributor·

  ·
  3

• ci: proxy CI run for PR #4657

  #4664·HuiyingLi opened Nov 6, 2026Contributor·

  ·
  3

• feat(data): integrate Zephon text datasets with training recipes

  community-request
  #4663·MaxiBoether opened Nov 6, 2026Contributor·3 of 4 tasks·

  ·
  1

• refactor(flow_matching): annotate timestep utilities and track with ty

  community-request
  #4661·Alan-donk opened Nov 6, 2026·4 tasks done·

  ·
  3

• fix(cuda_graphs): disable AOT donated buffers during partial graph capture

  bugQA_fixqa_rcca_doneregression
  #4660·pruprakash opened Nov 5, 2026Contributor·2 of 3 tasks·

  ·
  4

  

• chore: CI proxy for #4551

  #4658·akoumpa opened Nov 5, 2026Contributor·

  ·
  2

• feat(peft): NVFP4-resident grouped experts (expert_weight_format: nvfp4)

  community-request
  #4657·mreso opened Nov 5, 2026Contributor·3 tasks done·

  ·
  30

• refactor(training): annotate GarbageCollection and add unit tests

  community-request
  #4656·taran-dev4u opened Nov 5, 2026Contributor·3 tasks done·

  ·
  2

• chore(types): tighten distributed config scope annotations

  community-request
  #4655·wanjinhao1 opened Nov 5, 2026·3 tasks done·

  ·
  4

• fix(quantization): type QLoRA config and require load flags

  community-request
  #4654·sysuTobo opened Nov 5, 2026·

  ·
  3

• chore: CI proxy for #4647

  #4652·akoumpa opened Nov 5, 2026Contributor·

  ·
  3

• feat(moe): add DeepEP v2.5 expert-parallel dispatcher

  #4646·HuiyingLi opened Nov 4, 2026Contributor·3 tasks done·

  ·
  1

• fix(datasets): find answer start by token prefix in format_prompt_completion

  community-requestwaiting-on-maintainers
  #4643·spped2000 opened Nov 4, 2026·2 of 3 tasks·

  ·
  1

• chore: CI proxy for #4549

  #4639·akoumpa opened Nov 3, 2026Contributor·

  ·
  5

• refactor(distributed): keep TP trainability and drop reapply_trainability

  #4638·akoumpa opened Nov 3, 2026Contributor·3 tasks done·

  ·
  2

• feat(ci): add ci.perf_testing section to example configs

  #4637·nv-tjakubek opened Nov 3, 2026Contributor·

  ·
  4

• feat(loggers): add sound type annotations to log_utils

  community-request
  #4636·Goldloli opened Nov 3, 2026·

  ·
  2

• feat(minimax_m3_vl): train M3.1 DSpark drafts in the released layout

  community-request
  #4635·ischencheng opened Nov 3, 2026Contributor·3 tasks done·

  ·
  20

• fix(launcher): sound types for the interactive recipe boundary

  community-request
  #4633·zahidSkywalker opened Nov 3, 2026·

  ·
  4

  

• fix(moe): pass dispatcher_equal_token_counts through the TE experts

  #4629·HuiyingLi opened Nov 3, 2026Contributor·3 tasks done·

  ·
  3

• feat(utils): add type annotations to yaml_utils

  community-request
  #4628·AndreIsraelAO opened Nov 3, 2026·

  ·
  12