Issues · NVIDIA-NeMo/RL · GitHub

[meta] Before posting an issue
#701 · terrykong opened on Jul 21, 2025
[NeMo RL] v0.7.0 Release Roadmap
#2591 · anwithk opened on May 27, 2026
1

Labels Milestones New issue

[AGENT-2] Contributor skills lack user-facing troubleshooting and experiment-launcher coverage

#2729

· anwithk opened

on Jun 5, 2026

[AGENT-1] Add AGENTS.md symlink for discoverability across coding agents

#2728

· anwithk opened

on Jun 5, 2026

[DOCS-1] LoRA needs a dedicated page with backend-specific examples and schema comparison

#2727

· anwithk opened

on Jun 5, 2026

[TRAIN-10] Quickstart has no step cap; large dataset prep makes first run slow

#2726

· anwithk opened

on Jun 5, 2026

[TRAIN-9] Thinking models (Qwen3/Qwen3.5) silently produce 0% accuracy due to low max_new_tokens

#2725

· anwithk opened

on Jun 5, 2026

[TRAIN-8] No end-to-end tutorial for writing and registering custom reward functions

#2724

· anwithk opened

on Jun 5, 2026

[TRAIN-7] RuntimeError: No CUDA GPUs are available on A100; fix requires TORCH_CUDA_ARCH_LIST

#2723

· anwithk opened

on Jun 5, 2026

[TRAIN-6] flash-linear-attention and causal-conv1d missing from container and docs for Qwen3.5

#2722

· anwithk opened

on Jun 5, 2026

[TRAIN-5] Example configs are too fragmented; no canonical config per model family

#2721

· anwithk opened

on Jun 5, 2026

[TRAIN-4] AIME 2025 cannot be used directly as a validation dataset during training

#2720

· anwithk opened

on Jun 5, 2026

[TRAIN-3] Hydra override path data.dataset_name is incorrect; should be data.train.dataset_name

#2719

· anwithk opened

on Jun 5, 2026

[TRAIN-2] DTensor async GRPO with dis-colocated generation hangs when tensor_parallel=2

#2718

· anwithk opened

on Jun 5, 2026