Code Generation on HumanEval (Pass@1, 10, 100)
90.9Pass@1Teacher
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TeacherModel=Qwen2.5-7B-Instruct, Role=Teacher2025.10 | 90.9 | — | — | |
| DistiLLM-2alpha=-5, Base Model=Qwen2.5-1.5B-Instruct2025.10 | 73.2 | — | — | |
| DistiLLM-2alpha=-1, Base Model=Qwen2.5-1.5B-Instruct2025.10 | 72 | — | — | |
| StudentModel=Qwen2.5-1.5B-Instruct, Role=Student2025.10 | 70.7 | — | — | |
| PASERPruning Scheme=LLM-Pruner (25%)2025.02 | 14.4 | 27.6 | 48.2 | |
| PASERPruning Scheme=SliceGPT (25%)2025.02 | 12.9 | 25.2 | 44.5 | |
| NuggetsPruning Scheme=LLM-Pruner (25%)2025.02 | 11.8 | 22.9 | 38.3 | |
| IFDPruning Scheme=LLM-Pruner (25%)2025.02 | 10.5 | 21.2 | 35.6 | |
| NuggetsPruning Scheme=SliceGPT (25%)2025.02 | 9.5 | 18.5 | 33.9 | |
| Instruction MiningPruning Scheme=LLM-Pruner (25%)2025.02 | 8.9 | 17.8 | 28.4 | |
| IFDPruning Scheme=SliceGPT (25%)2025.02 | 8.7 | 16.8 | 31.2 | |
| Full DataPruning Scheme=LLM-Pruner (25%)2025.02 | 7.8 | 15.1 | 19 | |
| Instruction MiningPruning Scheme=SliceGPT (25%)2025.02 | 6.3 | 11.4 | 23.8 | |
| RandomPruning Scheme=LLM-Pruner (25%)2025.02 | 6.2 | 13.7 | 20.5 | |
| w/o TrainingPruning Scheme=LLM-Pruner (25%)2025.02 | 3.4 | 6.2 | 10.4 | |
| RandomPruning Scheme=SliceGPT (25%)2025.02 | 3.1 | 5.8 | 14.2 | |
| w/o TrainingPruning Scheme=SliceGPT (25%)2025.02 | 3 | 5.9 | 8.7 | |
| Full DataPruning Scheme=SliceGPT (25%)2025.02 | 2.9 | 5.1 | 11.8 |