Community Bot

Workflow runs for Community Bot

• chore: AUT-2539 nightly sync main into dev (2026-10-13)

  completed successfully:
  Community Bot#32359:Issue comment#8209 (comment)createdbysvcnvidia-nemo-ci
  Oct 16, 202619s
  More actions for workflow run 32359

• test(moe): refresh goldens for the DeepSeek proxy optimizer CUDA graph test

  completed successfully:
  Community Bot#32358:Issue comment#8240 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202616s
  More actions for workflow run 32358

• fix(gtp): preserve FP32 reduce-scatter outputs until gradient accumulation

  completed successfully:
  Community Bot#32357:Issue comment#7889 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202617s
  More actions for workflow run 32357

• test(moe): refresh goldens for the DeepSeek proxy optimizer CUDA graph test

  completed successfully:
  Community Bot#32356:Issue comment#8240 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202615s
  More actions for workflow run 32356

• [Dev] Add KPool DSA indexer, NoPE MLA support, and FP8 precision controls for the indexer and MLA attention (dev port of #7522)

  completed successfully:
  Community Bot#32355:Issue comment#8287 (comment)createdbybuptzyb
  Oct 16, 202615s
  More actions for workflow run 32355

• [Dev] Add KPool DSA indexer, NoPE MLA support, and FP8 precision controls for the indexer and MLA attention (dev port of #7522)

  completed successfully:
  Community Bot#32354:Issue comment#8287 (comment)createdbycopy-pr-bot[bot]Bot
  Oct 16, 202613s
  More actions for workflow run 32354

• Honor per-module quantization override in the fused GroupedMLP path

  completed successfully:
  Community Bot#32353:Issue comment#8286 (comment)createdbycopy-pr-bot[bot]Bot
  Oct 16, 202617s
  More actions for workflow run 32353

• [dev] Mask padding before HybridEP dispatch under an expert-rank budget

  completed successfully:
  Community Bot#32352:Issue comment#8285 (comment)createdbycopy-pr-bot[bot]Bot
  Oct 16, 202613s
  More actions for workflow run 32352

• fix(inference): AUT-2539 reply to a KV handoff import that fails before admission

  completed successfully:
  Community Bot#32351:Issue comment#8255 (comment)createdbysvcnvidia-nemo-ci
  Oct 16, 202612s
  More actions for workflow run 32351

• fix(inference): AUT-2539 hold back unstable reasoning in streamed chat deltas

  completed successfully:
  Community Bot#32350:Issue comment#8254 (comment)createdbysvcnvidia-nemo-ci
  Oct 16, 202617s
  More actions for workflow run 32350

• [dev] Share CUDA graph slots across dynamic CP variants

  completed successfully:
  Community Bot#32349:Issue comment#6479 (comment)createdbycopy-pr-bot[bot]Bot
  Oct 16, 202614s
  More actions for workflow run 32349

• [MFSDP v2] Record iteration demand order for VPP and combined 1F1B

  completed successfully:
  Community Bot#32348:Issue comment#8284 (comment)createdbycopy-pr-bot[bot]Bot
  Oct 16, 202617s
  More actions for workflow run 32348

• Perf and memory overheads fix for MTP and Cross Entropy

  completed successfully:
  Community Bot#32347:Issue comment#7478 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202612s
  More actions for workflow run 32347

• Fix CG + CP until TransformerEngine#3205 lands

  completed successfully:
  Community Bot#32346:Issue comment#8124 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202615s
  More actions for workflow run 32346

• Add per-head Muon QKV orthogonalization and fix TP>num_query_groups with Muon. Generalize to MLA and GQA

  completed successfully:
  Community Bot#32345:Issue comment#6326 (comment)createdbymkhona-nvidia
  Oct 16, 202615s
  More actions for workflow run 32345

• fix(inference): AUT-2539 fail impossible KV handoffs on arrival and honor aborts

  completed successfully:
  Community Bot#32344:Issue comment#8267 (comment)createdbysvcnvidia-nemo-ci
  Oct 16, 202615s
  More actions for workflow run 32344

• Fix CG + CP until TransformerEngine#3205 lands

  completed successfully:
  Community Bot#32343:Issue comment#8124 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202618s
  More actions for workflow run 32343

• Use weak-key membership for MFSDP context roots

  completed successfully:
  Community Bot#32342:Issue comment#8222 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202614s
  More actions for workflow run 32342

• refactor(training): migrate SchedulerConfig

  completed successfully:
  Community Bot#32341:Issue comment#8211 (comment)createdbycuichenx
  Oct 16, 202614s
  More actions for workflow run 32341

• test(optimizer): cover mixed-dtype checkpoint state and FP32 mHC

  completed successfully:
  Community Bot#32340:Issue comment#7804 (comment)createdbycuichenx
  Oct 16, 202613s
  More actions for workflow run 32340

• Pad the scale buffers before calling TE's quantize

  completed successfully:
  Community Bot#32339:Issue comment#8247 (comment)createdbyXiaodaNV
  Oct 16, 202613s
  More actions for workflow run 32339

• dist_ckpt: inference-only streaming checkpoint dequantization for --fp8-param-gather / --fp4-param-gather loads

  completed successfully:
  Community Bot#32338:Issue comment#7405 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202613s
  More actions for workflow run 32338

• Support NCCL UBR gradient-buffer offload in RL

  completed successfully:
  Community Bot#32337:Issue comment#6925 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202616s
  More actions for workflow run 32337

• Support NCCL UBR gradient-buffer offload in RL

  completed successfully:
  Community Bot#32336:Issue comment#6925 (comment)createdbynemo-automation-bot[bot]Bot
  Oct 16, 202615s
  More actions for workflow run 32336

• [Dev] Release MoE dispatcher routing tensors after combine

  completed successfully:
  Community Bot#32335:Issue comment#8231 (comment)createdbyyaox12
  Oct 16, 202616s
  More actions for workflow run 32335