Reasoning on ARC, StraQA, CSQA Overall
69.85AccuracyHSIR-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HSIR-SFTBackbone=Qwen2.5-1.5B, Training Iterations=12026.05 | 69.85 | 1,043 | |
| SFT-OracleBackbone=Qwen2.5-1.5B, Training Iterations=12026.05 | 68.14 | 1,333 | |
| ReSTEMBackbone=Qwen2.5-1.5B, Training Iterations=12026.05 | 66.74 | 1,254 | |
| STaRBackbone=Qwen2.5-1.5B, Training Iterations=12026.05 | 64.2 | 1,357 | |
| SFT-InitialBackbone=Qwen2.5-1.5B2026.05 | 62.49 | 1,447 |