There's the OIG dataset (https://laion.ai/blog/oig-dataset/ https://laion.ai/blog/oig-dataset/) which was used to train a NeoX 20B ChatBot (https://huggingface.

There's the OIG dataset (https://laion.ai/blog/oig-dataset/ https://laion.ai/blog/oig-dataset/) which was used to train a NeoX 20B ChatBot (https://huggingface.co/togethercomputer/GPT-NeoXT-Chat-Base-20B https://huggingface.co/togethercomputer/GPT-NeoXT-Chat-Base-...). The dataset is larger and publicly available. I want to try finetuning LLaMa on this tonight.