Proficiency Control on TinyTolkien
15Ctrl ErrorCALM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CALMdecoding_strategy=top-32024.06 | 15 | 9.3 | 9.7 | 330 | |
| CALMbackbone=LLama2-7b, training=Finetuned + PPO2024.06 | 39 | 9.2 | 9.6 | 110 | |
| Mistral-7btraining=Finetuned + PPO2024.06 | 60 | 9.1 | 9.7 | 110 | |
| Mistral-7btraining=Finetuned2024.06 | 69 | 9.4 | 9.9 | 110 | |
| LLama2-7btraining=Finetuned2024.06 | 81 | 9.3 | 9.8 | 110 |