Data Analysis on DACO (TestH)
43.92HelpfulnessGPT-4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4# para.=175B+, Code generation=true2024.03 | 43.92 | 3.26 | 17.54 | |
| ChatGPT# para.=20B+, Code generation=true2024.03 | 21.38 | 2.59 | 14.51 | |
| GPT-4# para.=175B+, Code generation=false2024.03 | 20.5 | 4.36 | 13.71 | |
| TAPAS# para.=337M, Code generation=false2024.03 | 16.5 | 3.67 | 9.73 | |
| ChatGPT# para.=20B+, Code generation=false2024.03 | 13.5 | 2.07 | 13.51 | |
| FG-RLHF# para.=6B, Code generation=true2024.03 | 12.5 | 5.98 | 11.8 | |
| SFT# para.=6B, Code generation=false2024.03 | 11.33 | 2.65 | 13.63 | |
| SFT# para.=6B, Code generation=true2024.03 | 9.83 | 4.47 | 14.6 | |
| TAPEX# para.=406M, Code generation=false2024.03 | 9 | 3.5 | 13.81 | |
| RLHF# para.=6B, Code generation=true2024.03 | 7.51 | 3.13 | 11.46 |