Code Generation on HumanEval+, MBPP+, LCB
85.2HumanEval+ ScoreTeacher
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TeacherModel Type=Teacher (RL), Teacher Models=Qwen3-4B-Math-RL and Qwen3-4B-Code-RL2026.05 | 85.2 | 69.8 | 26.6 | 60.5 | |
| Uni-OPDStudent Model=Qwen3-4B2026.05 | 83.1 | 71.3 | 28 | 60.8 | |
| OPDStudent Model=Qwen3-4B2026.05 | 82.9 | 72.4 | 21.6 | 59 | |
| TeacherModel=Qwen3-30B-A3B-Instruct-25072026.05 | 81.9 | 77.2 | 23.4 | 60.8 | |
| StudentStudent Model=Qwen3-4B2026.05 | 77.4 | 65.3 | 17.7 | 53.5 | |
| OPDDistillation Setting=Multi-Teacher Distillation2026.05 | 73.2 | 59.1 | 25.7 | 52.7 | |
| Uni-OPDDistillation Setting=Single-Teacher Distillation2026.05 | 73 | 60 | 28.1 | 53.7 | |
| Uni-OPDDistillation Setting=Multi-Teacher Distillation2026.05 | 72.9 | 60.5 | 28 | 53.8 | |
| OPDDistillation Setting=Single-Teacher Distillation2026.05 | 71.8 | 58.2 | 26.7 | 52.5 | |
| Uni-OPDStudent Model=Qwen3-1.7B2026.05 | 71.5 | 58.6 | 28 | 52.7 | |
| OPDStudent Model=Qwen3-1.7B2026.05 | 67.1 | 56.7 | 23.4 | 49.1 | |
| StudentModel Type=Qwen3-1.7B Student2026.05 | 61.9 | 53.4 | 11.9 | 42.4 | |
| StudentStudent Model=Qwen3-1.7B2026.05 | 61.9 | 53.4 | 11.9 | 42.4 |