Instruction Following on IFBench (pass@1, pass@128)
27Pass@1Base
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BaseBackbone=Qwen2.5-7B, Generation budget=4096 tokens2026.04 | 27 | 58 | |
| GRPO (Countdown)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Optimization=GRPO2026.04 | 27 | 61 | |
| Countdown (A) + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A), Optimization=GRPO2026.04 | 24 | 55 | |
| Countdown (A)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A)2026.04 | 23 | 54 | |
| Countdown (A') + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A'), Optimization=GRPO2026.04 | 20 | 58 | |
| Countdown (A')Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A')2026.04 | 19 | 62 |