Dialogue Generation on Human-in-the-loop Interactive Evaluation Customer-Agent Dialogs
41Win Rate (vs GPT-4)GER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GERStudent Model Architecture=LLaMA-2-13b2024.08 | 41 | 16 | |
| GERStudent Model Architecture=LLaMA-2-70b2024.08 | 41 | 3 | |
| GERStudent Model Architecture=GPT-3.52024.08 | 41 | 19 | |
| Base LLMStudent Model Architecture=LLaMA-2-70b2024.08 | 38 | — | |
| GERStudent Model Architecture=LLaMA-2-7b2024.08 | 34 | 31 | |
| Base LLMStudent Model Architecture=LLaMA-2-13b2024.08 | 25 | — | |
| Base LLMStudent Model Architecture=GPT-3.52024.08 | 22 | — | |
| Base LLMStudent Model Architecture=LLaMA-2-7b2024.08 | 3 | — |