| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Llama-3.3-70B math n=30 (failure pool) | L_MEMORY (Source-Conditioned Role Relabeling) | Correction Rate86.7 | 5 | 1mo ago | |
| Qwen-72B math failure pool n=30 | L_MEMORY (Source-Conditioned Role Relabeling) | Correction Rate80 | 5 | 1mo ago | |
| WikiText-2 and OpenWebText | Ours-0.6B | NLI Score82.2 | 3 | 2mo ago |