FreeWilly2 is an open-source AI model developed by Stability AI and fine-tuned with the Llama2 70B dataset.

It is a language model that uses auto-regressive techniques to generate text.

The model is trained using supervised fine-tuning on an internal Orca-style dataset.

Users can interact with FreeWilly2 by starting a conversation using a specific prompt format, which includes a system prompt, a user prompt, and an assistant response.

The model can generate responses based on the given prompts.

FreeWilly2 is implemented using the HuggingFace Transformers library and is available under the Non-Commercial Creative Commons license (CC BY-NC-4.0).

It is intended for research purposes only.

However, it's important to note that while the Llama2 dataset helps mitigate biases and toxicity in the generated text, not all biases can be eliminated through fine-tuning.

Users should be aware of potential issues and not treat the model outputs as a substitute for human judgment or a source of truth.

Responsible usage is strongly encouraged.

For further information or inquiries about the model, users can contact Stability AI via email at [email protected] .

Citations for the Llama2 dataset and the Orca-style training methodology are provided for academic references.