Mathematical Reasoning on Math500 (Pass@1, #L)
80.4Pass@1Latent-GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Latent-GRPOBase Model=Qwen2.5-Math-7B, Reasoning Strategy=Latent Reasoning, Sampling Mode=No Sampling2026.04 | 80.4 | 929 | |
| GRPOBase Model=Qwen2.5-Math-7B, Reasoning Strategy=Explicit Reasoning2026.04 | 75.15 | 2,982 | |
| SFTBase Model=Qwen2.5-Math-7B, Reasoning Strategy=Explicit Reasoning2026.04 | 67.4 | 2,662 | |
| Latent-SFTBase Model=Qwen2.5-Math-7B, Reasoning Strategy=Latent Reasoning, Sampling Mode=No Sampling2026.04 | 64.85 | 1,070 | |
| Soft-GRPOBase Model=Qwen2.5-Math-7B, Reasoning Strategy=Latent Reasoning, Sampling Mode=No Sampling2026.04 | 60.6 | 1,133 |