Open Monohydroxides opened 5 months ago
Following up on this, it seems the only perplexity metric in the repo is using the SEDD predictions as GT and measuring the perplexity of a pre-trained GPT-2 model, presumably to measure alignment with that model.
Would it be at all possible to provide the perplexity metric used in the paper?
Thanks!
Hello, thank you for your outstanding work! In the code, I don't seem to find the implementation for calculating likelihood values as given by equation (10) in the paper. Could you please inform me about when you plan to upload it to the GitHub repository? Or perhaps I missed a part of the code that covers this calculation. Could you please assist me in identifying where the code for calculating likelihood values is located?