Nemotron Nano 30B GRPO LoRA — step 10

• Base: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
• LoRA: dim=128, alpha=512, exclude *out_proj*
• Checkpoint: GRPO global step 10
• Training: NeMo RL 0.6.0, 8×A100, recipe-prod

Eval note (Jul 16 Harbor 5-way)

• Step 10: best pipeline/tool reliability (wrote_any_file / pipeline_flow = 100%)
• Step 20: best held-out content quality (OQ 2.05, IA 1.35); preferred demo checkpoint