Figure 3 explanation
Muennighoff opened this issue · 2 comments
Muennighoff commented
Sorry, a few things are unclear to me about Fig 3:
- What is generated by the model? Everything starting from
<work>
? Including Answer? - Is it zero-shot like in the image?
- Is it a mistake from the model that it generates « & >?
- Does the model literally predict the output of the program (0.0051) or did you run the program externally and then add it to the image?
Thanks 🤗
RJT1990 commented
Thanks for the comment: for inference, it generates everything after including Answer. It is "zero-shot" in the sense you just need to prompt with at test-time.
Muennighoff commented
Interesting, so the calculation 0.0051 was not done using an external program, but the model did it itself!
Thanks 👍