All pull requests

Pull requests

• [Fix] Make checkpoint rotation safe on shared and node-local storage

  #2183·cangtianhuang opened Oct 1, 2026Contributor·

  ·
  1

• [release/0.4] Support W4A8 (FP4 weight x FP8 activation) for the MoE expert GEMM

  #2182·chenbong opened Oct 1, 2026·

  ·
  7

• refactor: extract node timeout helper

  #2180·Ray961123 opened Oct 1, 2026Contributor·

  ·
  3

• [New features] Backward comm-compute overlap on the explicit ring driver

  #2179·adam-xiaoyao opened Sep 30, 2026Contributor·

  ·
  11

• [CI] Make Test-release work for stacked PRs

  #2177·zrr1999 opened Sep 30, 2026Member·1/11·

  ·

• [HyperBody] Unify token-averaged loss display and make pad_token_id configurable

  #2176·Lmywl opened Sep 30, 2026Contributor·

  ·
  6

• [CI] Use Paddle-Review-Bot as the only bot gate approver

  #2173·Paddle-Review-Bot opened Sep 30, 2026·

  ·
  1

• [CI] Use Paddle-Review-Bot as the only bot gate approver

  #2172·Paddle-Review-Bot opened Sep 30, 2026·

  ·
  6

• 清理CI的pip缓存

  #2171·XieYunshen opened Sep 30, 2026Collaborator·

  ·
  1

• [BugFix] Prevent cached wheel corruption from failing GPU CI

  #2170·XieYunshen with CopilotAI opened Sep 30, 2026·

  ·
  1

  

• docs: add deployment network notes

  #2169·Ray961123 opened Sep 30, 2026Contributor·

  ·
  3

• [FlexCheckpoint] Decouple empty-layer naming and dispatch AOA consumers

  #2168·wangbingguang2026 opened Sep 29, 2026Contributor·

  ·
  7

• [internal_medicine] Absorb grad_health monitor from llm-internal-medicine (upstream PR 72/73)

  #2167·yuweivvv opened Sep 29, 2026Contributor·

  ·
  3

• [FlexCheckpoint] Add modular AOA leaves for attention and MLA/CSA/DSA

  #2166·wangbingguang2026 opened Sep 29, 2026Contributor·

  ·
  6

• Build EMA HF checkpoint from shared memory instead of disk files

  #2165·cszdrg opened Sep 29, 2026Contributor·

  ·
  3

• refactor: extract node timeout helper

  #2162·Ray961123 opened Sep 29, 2026Contributor·

  ·
  2

• [gpt] fix int32 overflow in fill_feature at long context

  #2160·From00 opened Sep 28, 2026Collaborator·

  ·
  3

• [Checkpoint] Fix MTP flex_checkpoint export and magic-send AOA mapping

  #2157·zrr1999 opened Sep 28, 2026Member·2/11·

  ·
  1

• refactor: extract node timeout helper

  #2153·Ray961123 opened Sep 28, 2026Contributor·

  ·
  2

• docs: add deployment network notes

  #2152·Ray961123 opened Sep 27, 2026Contributor·

  ·
  2

• [erndata] Support context_parallel_size>1 without MTP (mtp0) under use_erndata

  #2151·cangtianhuang opened Sep 26, 2026Contributor·

  ·
  8

• docs: add deployment network notes

  #2150·Ray961123 opened Sep 26, 2026Contributor·

  ·
  2

• feat: add emergency response log tool

  #2149·Ray961123 opened Sep 25, 2026Contributor·

  ·
  2

• Pr 2086 review fix

  #2147·swgu98 opened Sep 24, 2026Member·

  ·
  3

• test: port #2080 coverage_test migration/rewrite to release/0.4

  #2146·liuhao2638 opened Sep 24, 2026Contributor·

  ·
  6