EDA on Celebrities-30 EDA (subsampled)
19.7#TurnsVicuna
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VicunaModel size=7B2023.10 | 19.7 | 5 | 6.2 | 0.04 | |
| Vicuna 7BSize=7B, Version=v1.32023.10 | 19.6 | 6 | 5.9 | 0.05 | |
| V-RLGPBackbone=Vicuna 7B2023.10 | 19.5 | 9 | 5.8 | 0.07 | |
| V-FT 7B (All)Size=7B, Training Data=All turns2023.10 | 19.3 | 16 | 7.6 | 0.13 | |
| V-FTBackbone=Vicuna 7B, Training Data=Things2023.10 | 19.1 | 19 | 10.3 | 0.15 | |
| V-FTBackbone=Vicuna 7B, Training Data=Celebs2023.10 | 19.1 | 20 | 7.5 | 0.16 | |
| V-FTBackbone=Vicuna 7B, Training Data=All2023.10 | 19 | 21 | 9.1 | 0.16 | |
| V-FT 7B (Suc.)Size=7B, Training Data=Successful turns only2023.10 | 19 | 15 | 6.3 | 0.13 | |
| GPT-3.52023.10 | 18.8 | 27 | 7.4 | 0.21 | |
| V-FT 13B (All)Size=13B, Training Data=All turns2023.10 | 18.8 | 22 | 6.2 | 0.18 | |
| V-RLGP 7BSize=7B2023.10 | 18.8 | 16 | 5.9 | 0.14 | |
| Vicuna 13BSize=13B, Version=v1.32023.10 | 18.7 | 22 | 6.1 | 0.18 | |
| V-FTBackbone=Vicuna 7B, Training Data=Success trajectories2023.10 | 18.6 | 21 | 7.4 | 0.17 | |
| V-FTBackbone=Vicuna 13B, Training Data=Success trajectories2023.10 | 18.6 | 22 | 7.8 | 0.18 | |
| V-RLGP 13BSize=13B2023.10 | 18.5 | 26 | 6.1 | 0.21 | |
| V-FT 13B (Suc.)Size=13B, Training Data=Successful turns only2023.10 | 18.4 | 23 | 5.9 | 0.19 | |
| Mistral 7BSize=7B2023.10 | 18.2 | 22 | 4.3 | 0.2 | |
| Human2023.10 | 18.1 | 31 | 7 | 0.25 | |
| GPT-3.52023.10 | 17.9 | 41 | 7.6 | 0.33 | |
| VicunaModel size=13B2023.10 | 17.7 | 36 | 6.8 | 0.27 | |
| Claude-12023.10 | 17.7 | 29 | 5.3 | 0.25 | |
| Claude-22023.10 | 17.6 | 31 | 5.6 | 0.26 | |
| V-RLGPBackbone=Vicuna 13B2023.10 | 17.5 | 35 | 6.8 | 0.29 | |
| GPT-42023.10 | 17.3 | 50 | 6.8 | 0.4 | |
| Claude-12023.10 | 16.7 | 41 | 4.6 | 0.35 | |
| GPT-42023.10 | 16.5 | 59 | 7.3 | 0.48 | |
| Claude-22023.10 | 15.9 | 45 | 5.3 | 0.4 |