Description
This dataset is a filtered and simplified version of the original tatsu-lab/alpaca dataset. It is designed specifically for Reinforcement Learning (RL) stages where only a prompt is required for model generation.
Processing Steps
Filtering: All rows containing values in the input column were removed. This ensures the dataset only contains standalone instructions that do not require additional context.
Column Reduction: All columns except for instruction were removed.
Shuffling: The… See the full description on the dataset page:
https://huggingface.co/datasets/Seungjun/toxic_llama_v2_cleaned_alpaca.
