Scientific Reasoning on GPQA Extended (Pass@1)
31.1Pass@1 AccuracyMathstral-7B-v0.1 + GRPO w/ VERL.
Evaluation results
| Method | Links | |
|---|---|---|
| Mathstral-7B-v0.1 + GRPO w/ VERL.Base Model=Mathstral-7B-v0.1, Training=GRPO w/ VERL.2025.09 | 31.1 | |
| Qwen2.5-7B + PPO w/ VERL.Base Model=Qwen2.5-7B, Training=PPO w/ VERL.2025.09 | 30 | |
| Mathstral-7B-v0.1 + PPOBase Model=Mathstral-7B-v0.1, Training=PPO2025.09 | 29.9 | |
| Mathstral-7B-v0.1 + GRPOBase Model=Mathstral-7B-v0.1, Training=GRPO2025.09 | 29.3 | |
| Qwen2.5-7B + GRPO w/ VERL.Base Model=Qwen2.5-7B, Training=GRPO w/ VERL.2025.09 | 28.9 | |
| Qwen2.5-7B + GRPOBase Model=Qwen2.5-7B, Training=GRPO2025.09 | 27.8 | |
| Mathstral-7B-v0.1 + PPO w/ VERL.Base Model=Mathstral-7B-v0.1, Training=PPO w/ VERL.2025.09 | 26.6 | |
| Qwen2.5-7BBase Model=Qwen2.5-7B2025.09 | 26.6 | |
| Mathstral-7B-v0.1Base Model=Mathstral-7B-v0.12025.09 | 26 | |
| Qwen2.5-7B + PPOBase Model=Qwen2.5-7B, Training=PPO2025.09 | 25.3 |