Major bug: Chat template is not actually applied in run_sft.py and run_dpo.py

[AlexiaJM]

In run_sft.py and run_dpo.py, it says that it applies the chat template. But this is not actually done.

In the code below, column_names contains all the names of the columns, and this is given as a remove_columns argument to the mpa function. This means that all columns are skipped and the map is not applied at all. I actually tested by adding prints inside the apply_chat_template and I observed no printout at all, unless I removed the "remove_columns=column_names" in which case I was able to see printouts. Thus, the map is not applied at all and I wonder how it can even work.

column_names = list(raw_datasets["train"].features) ##################### # Apply chat template ##################### raw_datasets = raw_datasets.map( apply_chat_template, fn_kwargs={"tokenizer": tokenizer, "task": "dpo"}, num_proc=data_args.preprocessing_num_workers, remove_columns=column_names, desc="Formatting comparisons with prompt template", )