Improve Model by Using Better Dataset

#3
by isr431 - opened

I read your document about the technique you used to create these models. I also took a look at your dataset and noticed that it is surprisingly low quality and contains a lot of poor data. Would you be able to enhance the model's capabilities even further by using better datasets? I recommend checking the datasets created by these people/organizations: https://huggingface.co./NousResearch, https://huggingface.co./teknium, https://huggingface.co./datasets/cognitivecomputations/ and https://huggingface.co./anthracite-org. They should cover a wide variety of use cases and shouldn't have many refusals.

Sign up or log in to comment