Mathematical and General Reasoning on GSM8k + MATH 1.0 (test)
94.7GSM8k AccuracyBF16
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BF16Model=Qwen3-8B-Base, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 94.7 | 81.7 | 46.2 | 33.5 | 64 | |
| Jet-RLModel=Qwen3-8B-Base, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 92.1 | 76.7 | 43.3 | 33.2 | 61.3 | |
| BF16Model=Qwen2.5-7B, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 91.5 | 72.8 | 28.4 | 42.2 | 58.7 | |
| BF16-Train-FP8-RolloutModel=Qwen2.5-7B, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 90.7 | 68.1 | 28.9 | 26.9 | 53.7 | |
| Jet-RLModel=Qwen2.5-7B, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 89.3 | 69.2 | 36.5 | 27.8 | 55.7 | |
| Before TuningModel=Qwen3-8B-Base, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 63.7 | 69.7 | 45.7 | 30.7 | 52.5 | |
| Before TuningModel=Qwen2.5-7B, Rollout Length=16k, Training Dataset=GSM8k + MATH2026.01 | 13.5 | 45.3 | 28.4 | 26 | 28.3 |