All pull requests

Pull requests

• Gather write results as compact tables in async save finalize

  #446·shurkat-nvidia opened Oct 9, 2026Contributor·2/2·

  ·
  4

• Drop storage_data from cached Metadata

  #445·shurkat-nvidia opened Oct 9, 2026Contributor·1/2·

  ·
  2

• Use explicit health-check results for node decisions

  #442·namitdhameja opened Oct 9, 2026Contributor·

  ·
  2

• Add resilient external attrsvc lifecycle

  ci-approved
  #441·namitdhameja opened Oct 9, 2026Contributor·

  ·
  10

  

• #426 - L1 taxonomy: classify cancellations and cascade-only TCPStore logs

  #439·continue-revolution opened Oct 8, 2026Contributor·

  ·
  5

• Speed up checkpoint .metadata writes and reuse global metadata across restarts

  ci-approved
  #438·shurkat-nvidia opened Oct 8, 2026Contributor·

  ·
  4

• Start writer threads before the calling thread writes its bucket

  #436·luccabb opened Oct 5, 2026·

  ·
  2

• [WIP] Release non-cached GPU tensor references after async checkpoint staging

  #434·Elcfin opened Oct 5, 2026·

  ·
  1

  1

• Count unknown-progress attempts, not uncomparable prior attempts

  #431·continue-revolution opened Oct 1, 2026Contributor·

  ·
  2

• #384 - Fix straggler detection documentation import path

  #429·efegokdemir opened Sep 29, 2026·

  ·
  2

• Add Slack notifications for attribution results in nvrx-smonsvc

  #427·continue-revolution opened Sep 16, 2026Contributor·

  ·
  28

• [WIP] #405 - Optimize CPU-SHM snapshot copies

  #407·0z5a opened Aug 17, 2026·

  ·
  2

• Lens telemetry onto main

  #401·rhewett-nv opened Aug 10, 2026Contributor·

  ·
  7

• async_ckpt: tag cached identifiers with the worker generation

  #395·hexinw-nvidia opened Aug 5, 2026Contributor·

  ·
  1

• Feat/ckpt metadata reuse rebased

  #377·asolergi-nv opened Jul 22, 2026·

  ·
  2

• Add Restart Agent evaluation harness

  #374·namitdhameja opened Jul 14, 2026Contributor·

  ·
  

• Node-local checkpoint staging: async persist off the training critical path

  ci-approved
  #373·hexinw-nvidia opened Jul 13, 2026Contributor·

  ·
  18

• Enable progressive MCP log analysis

  ci-approved
  #368·namitdhameja opened Jul 2, 2026Contributor·

  ·
  2

  

• [codex] Bound LogSage endpoint outer retries

  ci-approved
  #367·namitdhameja opened Jul 1, 2026Contributor·

  ·
  2

  

• [codex] Recover exited managed attribution service

  ci-approved
  #361·namitdhameja opened Jun 29, 2026Contributor·

  ·
  4

  

• [codex] Fix LogSage LLM failure status mapping

  ci-approved
  #360·namitdhameja opened Jun 29, 2026Contributor·

  ·
  1

  

• Prepared-metadata reuse for collective-free async checkpoint saves

  ci-approved
  #359·asolergi-nv opened Jun 29, 2026·

  ·
  22

• Add scheduler exclusion producer service

  #358·namitdhameja opened Jun 27, 2026Contributor·

  ·
  11

  

• [codex] Filter stale FR dumps by Pstart time

  #355·namitdhameja opened Jun 24, 2026Contributor·

  ·
  4

  

• Persist async checkpoint plans across resume

  ci-approved
  #353·sbak5 opened Jun 23, 2026Contributor·

  ·
  6