What are Train Dataset labels ?

#1
by humza-sami - opened

I was going through train data while trying to reproduce the results. I am confused about the dataset. We are predicting the code block based on the query in the text. In the training dataset, there is content(code) but there is no query text on which model is trained. I am assuming the model is trained on (query, code) pair. How can I check the untokenzied queries for each example?

CodeParrot org

Hi, this dataset was used for model pertaining where the model predicts the next tokens of a sequence, so the labels are the same as the inputs (content column). You can find the training script here. If you want to fine-tune the model on text to code for example, you can find other datasets in this format such as CoNaLa and GitHub-Jupyter

loubnabnl changed discussion status to closed

Sign up or log in to comment