Text Fine Tuning Guidance

Helix uses the Mistral series of large language models to provide high quality responses with relatively small memory GPU footprint, including fine-tuning teach the model new information or styles of reasoning and presenting information.

Data Best Practices

• Your data must answer the question
• More data is better
• Only include data that is relevant to your problem

Use Case Best Practices

Question Answering

• Format training data as question-answer pairs
• Ensure your training data includes questions similar to downstream tasks
• Ensure question answer pairs have good “coverage” of your problem domain

Inference Best Practices

• Imagine yourself performing the task or answering the question
• If you need to ask questions, then you haven’t provided enough context
• If you are asking questions about a thing, indicate what the thing is

Reach out!

Please suggest improvements to this doc or encourage us to write more on this on Discord!