Terracotta is a free tool designed for fine-tuning and evaluating LLMs. Here is a blog post about why fine-tuning is important.
Here is what Terracotta can do:
- Training: Easily upload your dataset and fine-tune various OpenAI LLMs within minutes using our intuitive training dashboard. No more complex setup or coding—just focus on optimizing your models for your specific tasks.
- Qualitative Evaluation: Our playground allows you to compare prompting base models from OpenAI and Cohere against fine-tuned models, empowering you to assess the impact of your fine-tuning efforts and make informed decisions about model selection.
- Quantitative Evaluation: Experience our powerful evaluation tool that enables you to run inference on your dataset with any model in just a few clicks. Compare multiple models across different metrics relevant to your task, providing you with valuable insights and performance comparisons.
https://www.loom.com/share/da4ad333a5744f02852407997dfda181?sid=bc72296d-2a1e-4218-888a-9697c3870e74
There are two parts needed to run Terracotta: a React app and a Flask app. These are kept in separate folders
For the React app:
- Navigate to
nextjs_appdirectory - Do
npm installto install all the dependencies - Then, to run the frontend do
npm run dev
For the Flask app:
- Navigate to the
flask_appdirectory - Do
pip -r requirements.txtto install the dependencies - To run the app, do
flask run