Behavior Elicitation on Countdown (test)
87ScoreGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen3-8B2026.06 | 87 | |
| CPEBackbone=Qwen3-8B2026.06 | 85 | |
| GRPOBackbone=Llama-3.1-8B2026.06 | 82 | |
| SAEBackbone=Qwen3-8B2026.06 | 78 | |
| Random-LoRABackbone=Qwen3-8B2026.06 | 75 | |
| BaseBackbone=Qwen3-8B2026.06 | 73 | |
| CPEBackbone=Llama-3.1-8B2026.06 | 36 | |
| SAEBackbone=Llama-3.1-8B2026.06 | 32 | |
| Random-LoRABackbone=Llama-3.1-8B2026.06 | 8 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 7 |