It's not clear for me. I would like to create an AI chat bot for documentation, that is able to answer any questions about the topic.
#When to use fine tuning vs embeddings?
north bane
wraith crow
Embeddings allow you to create a similarity mapping between words, words given into embeddings endpoint are given back to you represented as vectors, and in your programming language of choice you can analyze these vectors to see what's similar to each other. This makes it so that you can create associations between phrases and other phrases and store them locally, and lookup the associations when you need
Fine tuning lets you put in prompt <> response pairs and train the model directly
the trained model will be saved and you can make requests to the trained model just like any of the regular flagship models
For your use case, either would work really, but using embeddings seems efficient and cost-effective