This dataset is a version of code_tasks_33k but modified to have the prompt be the incomplete code, and a request to provide what code should be entered in the middle.
The goal of this dataset is to train models that are highly capable at FIM (fill in the middle) and Tab-Autocomplete code tasks.
Since it uses the ShareGPT format, this uses a different method of autocomplete / FIM versus standard models. I have high hopes it will still work in applications such as Void, but I am unsure as of… See the full description on the dataset page:
https://huggingface.co/datasets/Sweaterdog/fim_code_tasks_33k.
