Softlabels aquired using Qwen2.5-1.5B-Instruct.
It is recommended to use this version of dataset when needed to train on softlabels: nixiieee/AG-news-softlabels-averaged.
The prompt was the following:
I have a news article and four possible categories it may belong to. Your task is to evaluate how likely the article belongs to each category, and return a probability (a number between 0 and 1, exclusive) for each one. The probabilities should sum approximately to 1, and reflect your confidence… See the full description on the dataset page:
https://huggingface.co/datasets/nixiieee/AG-news-softlabels.
