Logical Reasoning on LogicBench
80.4AccuracySA-GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SA-GRPOBackbone=Qwen3-4B2025.09 | 80.4 | 208 | |
| PPOBackbone=Qwen3-4B2025.09 | 80 | 455 | |
| GRPOBackbone=Qwen3-4B2025.09 | 78.4 | 416 | |
| SA-PPOBackbone=Qwen3-4B2025.09 | 77.6 | 342 | |
| GRPO-ERBackbone=Qwen3-4B2025.09 | 77.6 | 263 | |
| GRPO-O1Backbone=Qwen3-4B2025.09 | 76 | 350 | |
| BaseBackbone=Qwen3-4B2025.09 | 75.4 | 336 | |
| GRPOBackbone=Phi-3.5-mini2025.09 | 67.6 | 243 | |
| BaseBackbone=Phi-3.5-mini2025.09 | 66 | 309 | |
| SA-GRPOBackbone=Qwen3-1.7B2025.09 | 65 | 278 | |
| SA-PPOBackbone=Qwen3-1.7B2025.09 | 64.6 | 368 | |
| SA-PPOBackbone=Phi-3.5-mini2025.09 | 64.2 | 205 | |
| GRPOBackbone=Qwen3-1.7B2025.09 | 64.2 | 372 | |
| GRPO-ERBackbone=Qwen3-1.7B2025.09 | 61.4 | 206 | |
| PPOBackbone=Phi-3.5-mini2025.09 | 60.8 | 229 | |
| GRPO-O1Backbone=Qwen3-1.7B2025.09 | 60.6 | 264 | |
| PPOBackbone=Qwen3-1.7B2025.09 | 60.2 | 340 | |
| SA-GRPOBackbone=Phi-3.5-mini2025.09 | 57.4 | 173 | |
| BaseBackbone=Qwen3-1.7B2025.09 | 56 | 182 | |
| SA-PPOBackbone=Gemma3-1B2025.09 | 55.6 | 241 | |
| SA-GRPOBackbone=Gemma3-1B2025.09 | 55.2 | 222 | |
| GRPOBackbone=Gemma3-1B2025.09 | 53 | 277 | |
| GRPO-O1Backbone=Gemma3-1B2025.09 | 52.6 | 223 | |
| GRPO-O1Backbone=Phi-3.5-mini2025.09 | 51 | 212 | |
| BaseBackbone=Gemma3-1B2025.09 | 50.6 | 303 | |
| GRPO-ERBackbone=Gemma3-1B2025.09 | 50 | 250 | |
| GRPO-ERBackbone=Phi-3.5-mini2025.09 | 49.8 | 143 | |
| PPOBackbone=Gemma3-1B2025.09 | 48.8 | 264 |