General Language Proficiency on Aggregated (GSM8K, TruthfulQA, TriviaQA, CNN/DM, MMLU)
48.6Average ScoreLoRA Tuning
Evaluation Results
| Method | Links | |
|---|---|---|
| LoRA TuningModel Size=13B, Transfer Source=N/A2024.06 | 48.6 | |
| Full Fine-tuningModel Size=13B, Transfer Source=N/A2024.06 | 47.79 | |
| OursModel Size=13B, Transfer Source=7B2024.06 | 46.09 | |
| Proxy TuningModel Size=13B, Transfer Source=7B2024.06 | 44.43 | |
| Full Fine-tuningModel Size=13B, Transfer Source=7B2024.06 | 44.13 | |
| OursModel Size=13B, Transfer Source=1.1B2024.06 | 34.86 | |
| Proxy TuningModel Size=13B, Transfer Source=1.1B2024.06 | 31.74 | |
| Base ModelModel Size=13B, Transfer Source=N/A2024.06 | 29.93 | |
| Full Fine-tuningModel Size=13B, Transfer Source=1.1B2024.06 | 25.33 |