Rationale Generation on GSM8K (train)
98.9Rationale Validity RateTRICE without CV
Evaluation Results
| Method | Links | |
|---|---|---|
| TRICE without CVInitialization=Bootstrapped Q-R-A2023.11 | 98.9 | |
| TRICE with CVInitialization=Bootstrapped Q-R-A2023.11 | 98.8 | |
| TRICE with CV (not bootstrapped)Initialization=Random Q-R-A2023.11 | 98.4 | |
| STaRInitialization=Bootstrapped Q-R-A2023.11 | 80.2 |