Mathematical Reasoning on Olympiad (Accuracy @avg1)
0.457Accuracy (@avg1)LESS
Evaluation Results
| Method | Links | |
|---|---|---|
| LESSBackbone=Qwen2.5-Math-7B2025.11 | 0.457 | |
| GRPOBackbone=Qwen2.5-Math-7B2025.11 | 0.449 | |
| KL-CovBackbone=Qwen2.5-Math-7B2025.11 | 0.437 | |
| Forking TokensBackbone=Qwen2.5-Math-7B2025.11 | 0.422 | |
| GRPOBackbone=Qwen2.5-Base-7B2025.11 | 0.409 | |
| KL-CovBackbone=Qwen2.5-Base-7B2025.11 | 0.408 | |
| LESSBackbone=Qwen2.5-Base-7B2025.11 | 0.408 | |
| Forking TokensBackbone=Qwen2.5-Base-7B2025.11 | 0.406 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2025.11 | 0.396 | |
| LESSBackbone=Qwen2.5-Math-1.5B2025.11 | 0.396 | |
| Forking TokensBackbone=Qwen2.5-Math-1.5B2025.11 | 0.385 | |
| KL-CovBackbone=Qwen2.5-Math-1.5B2025.11 | 0.384 | |
| Base LLMBackbone=Qwen2.5-Math-7B2025.11 | 0.38 | |
| Base LLMBackbone=Qwen2.5-Math-1.5B2025.11 | 0.265 | |
| Base LLMBackbone=Qwen2.5-Base-7B2025.11 | 0.239 |