Issues

• Add per-checkpoint GPU memory checks to deployment tests

  NVIDIA/Model-Optimizer#3364 · Filter by author coderabbitai opened on Nov 12, 2026
  

• Tool-calling data in streaming training

  Enhancement

  NVIDIA/Model-Optimizer#3350 · Filter by author h-guo18 opened on Nov 11, 2026
  1

• Investigate: Megatron backend with vLLM streaming

  Enhancement

  NVIDIA/Model-Optimizer#3349 · Filter by author h-guo18 opened on Nov 11, 2026

• Investigate: linear-attention drafter

  Enhancement

  NVIDIA/Model-Optimizer#3348 · Filter by author h-guo18 opened on Nov 11, 2026

• Investigate: MoE drafter

  Enhancement

  NVIDIA/Model-Optimizer#3347 · Filter by author h-guo18 opened on Nov 11, 2026

• SpecDec Visualizer: UI refresh

  Enhancement

  NVIDIA/Model-Optimizer#3346 · Filter by author h-guo18 opened on Nov 11, 2026

• SpecDec Visualizer: data refresh and cleanup

  Task

  NVIDIA/Model-Optimizer#3345 · Filter by author h-guo18 opened on Nov 11, 2026

• Context/sequence parallelism for DFlash-family training

  Enhancement

  NVIDIA/Model-Optimizer#3344 · Filter by author h-guo18 opened on Nov 11, 2026
  1

• Unify DFlash-family training code

  Task

  NVIDIA/Model-Optimizer#3341 · Filter by author h-guo18 opened on Nov 11, 2026

• Qwen3-8B streaming DSpark example + walkthrough

  Enhancement

  NVIDIA/Model-Optimizer#3339 · Filter by author h-guo18 opened on Nov 11, 2026

• Cosmos3 Nano DFlash recipe as launcher pipeline

  Enhancement

  NVIDIA/Model-Optimizer#3338 · Filter by author h-guo18 opened on Nov 11, 2026

• Kimi-K3: loss mask, final norm, DSpark example

  Enhancement

  NVIDIA/Model-Optimizer#3337 · Filter by author h-guo18 opened on Nov 11, 2026

• Gemma-4-E4B as streaming DFlash/DSpark target

  Enhancement

  NVIDIA/Model-Optimizer#3336 · Filter by author h-guo18 opened on Nov 11, 2026

• Muse-Spark-Open DFlash2 recipes

  Enhancement

  NVIDIA/Model-Optimizer#3335 · Filter by author h-guo18 opened on Nov 11, 2026

• specdec_bench: fix run metadata, add MLflow

  Enhancement

  NVIDIA/Model-Optimizer#3330 · Filter by author h-guo18 opened on Nov 11, 2026

• Deprecate HF acceptance-rate validation

  Task

  NVIDIA/Model-Optimizer#3329 · Filter by author h-guo18 opened on Nov 11, 2026

• Offline acceptance-length eval on dumped hidden states

  Enhancement

  NVIDIA/Model-Optimizer#3328 · Filter by author h-guo18 opened on Nov 11, 2026

• Configurable loss terms; opt-in teacher softcap

  Enhancement

  NVIDIA/Model-Optimizer#3327 · Filter by author h-guo18 opened on Nov 11, 2026

• D-PACE position-aware loss (DFlash, DSpark, Domino)

  Enhancement

  NVIDIA/Model-Optimizer#3326 · Filter by author h-guo18 opened on Nov 11, 2026

• Apply recipe settings on DFlash resume

  Bug

  NVIDIA/Model-Optimizer#3318 · Filter by author h-guo18 opened on Nov 11, 2026

• Loss and mask fixes: anchor/selector weighting, padding, label shift

  Bug

  NVIDIA/Model-Optimizer#3316 · Filter by author h-guo18 opened on Nov 11, 2026

• Streaming reliability: vLLM 0.29 capture, corpus mmap, /done check

  Bug

  NVIDIA/Model-Optimizer#3315 · Filter by author h-guo18 opened on Nov 11, 2026

• Data echoing (reuse streamed samples)

  Enhancement

  NVIDIA/Model-Optimizer#3307 · Filter by author h-guo18 opened on Nov 11, 2026

• Fuse lm_head into draft loss

  Enhancement

  NVIDIA/Model-Optimizer#3306 · Filter by author h-guo18 opened on Nov 11, 2026

• [Spec Dec] Code Health & Refactoring

  roadmap
  Epic

  NVIDIA/Model-Optimizer#3303 · Filter by author h-guo18 opened on Nov 11, 2026