Summary

FAQs

• How can a fine-tuned model perform better?

  Base models like GPT4o are generalist, a fine-tuned model is a specialist. By fine-tuning, we train the model to focus primarily on our use-case, we can feed it as many examples as we want to use as a reference, leading to better performance.

• How can a fine-tuned model be cheaper?

  In general, with a fine-tuned model there is no need to prompt, leading to fewer tokens used. Fine-tuned models can be scaled down in size if needed, leading to up to 20x cheaper cost per tokens.

• How can a fine-tuned model be faster?

  Since we are using fewer tokens, and in some cases a much smaller model, the latency of fine-tuned models is very low compared to larger models.