You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
All parameters are self-explanatory, but `--sample_calc` means we will use calculator sampling. That is whenever an operation such as `<< 4 + 5=9 >>` is generated, we will invoke a calculator module to compute the result.
30
30
31
31
### Steps 2 and 3: Alignment and Discriminator Training
<palign="center"> <imgsrc="https://github.com/mukhal/grace/assets/5109053/ebbefdc2-0861-4fbc-ad0f-43316741bf58"alt="Description of the image"width="800"height="250"> </p>
33
33
34
34
Now we want to train a FLAN-T5 encoder as a discriminator over the sampled solutions.
0 commit comments