Math Reasoning on MATH500 (Accuracy and Delta %)
85AccuracyED-GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ED-GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 85 | 10.4 | |
| ETOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 84.8 | 10.2 | |
| DAPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 84.8 | 10.2 | |
| ED-iDPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 84.6 | 10 | |
| CoTBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 84.4 | 9.8 | |
| GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.05 | 83.8 | 9.2 | |
| ED-GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 77.6 | 3 | |
| ED-iDPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 77 | 2.4 | |
| GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 76.2 | 1.6 | |
| DAPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 75.8 | 1.2 | |
| ETOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 75.4 | 0.8 | |
| CoTBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.05 | 74.6 | — | |
| ED-iDPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 62 | 16.4 | |
| ETOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 61.6 | 16 | |
| ED-GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 61.4 | 15.8 | |
| GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 57.2 | 11.6 | |
| CoTBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 57 | 11.4 | |
| DAPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.05 | 54.4 | 8.8 | |
| ED-GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 49.2 | 3.6 | |
| ED-iDPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 47 | 1.4 | |
| ETOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 46 | 0.4 | |
| CoTBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 45.6 | — | |
| GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 45.4 | -0.2 | |
| DAPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.05 | 44.4 | -1.2 |