Data Analysis on DACO (test A)
50.79HelpfulnessGPT-4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4# para.=175B+, Code generation=true2024.03 | 50.79 | 4.59 | 17.77 | |
| GPT-4# para.=175B+, Code generation=false2024.03 | 30.43 | 3.35 | 14.9 | |
| ChatGPT# para.=20B+, Code generation=true2024.03 | 26.51 | 2.74 | 14.22 | |
| FG-RLHF# para.=6B, Code generation=true2024.03 | 19.42 | 3.65 | 13.13 | |
| ChatGPT# para.=20B+, Code generation=false2024.03 | 19.31 | 3.06 | 13.22 | |
| TAPAS# para.=337M, Code generation=false2024.03 | 19.19 | 1.96 | 11.62 | |
| SFT# para.=6B, Code generation=false2024.03 | 18.96 | 2.3 | 14.47 | |
| TAPEX# para.=406M, Code generation=false2024.03 | 15.08 | 3.34 | 14.6 | |
| SFT# para.=6B, Code generation=true2024.03 | 13.73 | 2.15 | 14.88 | |
| RLHF# para.=6B, Code generation=true2024.03 | 10.64 | 3.18 | 12.66 |