Language Understanding on MMLU-Pro (pass@k)
56pass@1Base
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaseBackbone=Qwen2.5-7B, Generation budget=4096 tokens2026.04 | 56 | 82 | |
| GRPO (Countdown)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Optimization=GRPO2026.04 | 56 | 84 | |
| Countdown (A') + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A'), Optimization=GRPO2026.04 | 52 | 93 | |
| Countdown (A) + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A), Optimization=GRPO2026.04 | 50 | 93 | |
| Countdown (A')Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A')2026.04 | 50 | 94 | |
| Countdown (A)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A)2026.04 | 46 | 91 |