MAAS GRPO Run

• model: Qwen/Qwen2-0.5B-Instruct
• epochs: 1
• batch size: 2
• gradient accumulation steps: 1
• num generations: 2
• learning rate: 5e-06
• dataset size: 7
• output dir: /workspace/out
• train metrics: {"total_flos": 0.0, "train_loss": 0.0, "train_runtime": 59.8793, "train_samples_per_second": 0.117, "train_steps_per_second": 0.117}