turboderp / Cat-Llama-3-70B-instruct

For each pulled record, we measure the perplexity of the entry against the gpt4 trained model, and select for specifically GPT-4 quality dataset.