Note Contains the first 1K rows of `nvidia/HelpSteer` without the original annotations, and using GPT-4 generated annotations instead for the categories: correctness, coherence, complexity, verbosity, and helpfulness.

Note Contains the first 1K rows of `nvidia/HelpSteer` without the original annotations, and using GPT-4 generated annotations instead for the categories: correctness, coherence, complexity, verbosity, and helpfulness. This dataset differs from the one above, as it contains also the raw responses and some other additional columns.

Note Dataset generated for the post at /blog/alvarobartt/replacing-judges-with-juries-distilabel