Coding on HumanEval (pass@1 and Overall Avg.)
57.3Pass@1Anchored Learning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Anchored LearningBackbone=Llama-3.2-3B-Instruct2026.05 | 57.3 | 52 | |
| Iter-SFTBackbone=Llama-3.2-3B-Instruct2026.05 | 56.4 | 39.6 | |
| BaseBackbone=Llama-3.2-3B-Instruct2026.05 | 54.9 | 39.4 | |
| Self-SFTBackbone=Llama-3.2-3B-Instruct2026.05 | 54.3 | 37.8 | |
| Low-SFTBackbone=Llama-3.2-3B-Instruct2026.05 | 53 | 45.6 | |
| STMBackbone=Llama-3.2-3B-Instruct2026.05 | 50.6 | 41.8 | |
| DFTBackbone=Llama-3.2-3B-Instruct2026.05 | 23.2 | 24.2 | |
| KL-SFTBackbone=Llama-3.2-3B-Instruct2026.05 | 18.3 | 20.7 | |
| SFTBackbone=Llama-3.2-3B-Instruct2026.05 | 6 | 19.1 |