Scientific Reasoning Tasks CoT Self-Instruct Starting Qwen3.5-4B model
80.22Average Score @8Agentic Self-Instruct
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Agentic Self-InstructTraining Data Source=Agentic Self-Instruct, Base Model=Qwen3.5-4B2026.06 | 80.22 | 94.36 | 3.05 | |
| Combined (2× data)Training Data Source=Combined, Training Data Size=2× data, Base Model=Qwen3.5-4B2026.06 | 79.66 | 93.5 | 2.49 | |
| CoT Self-InstructTraining Data Source=CoT Self-Instruct, Base Model=Qwen3.5-4B2026.06 | 79.03 | 92.77 | 1.86 | |
| Qwen3.5-4BTraining Data Source=Base, Base Model=Qwen3.5-4B2026.06 | 77.17 | 92.52 | — |