Can Model be trained from books?

Hey, so basically what I want to achieve is to train GPT-3 Model to be able to create questions from set of many books and articles.
Let’s say it will train on books about dinosaurs, then will create a set of questions based on learning from them. For example - which dinosaur is the tallest ?
Is this something that can be done ?

2 Likes

Hi Alex, If you can come up with simple prompts and completions, you can use a fine-tuning file

However, large documents are often processed using embedding. It will involve more work at your end - but it will ringfence the knowledge to what is in your articles and books

2 Likes

thanks ray, this is helpful. I need to figure out a source to get the embeddings so that I could train on my documents. can you point me to the direction pl

You need to break your documents down and use them as text for the embedding

There is a series of videos here that will explain the process

https://thoughtblogger.com/openai-embedding-tutorial/

9 Likes

OMG you are super fast, and big thanks. the provided link is exhaustive with example videos. will go through it soon

3 Likes

@ raymonddavey thank you very much for the fast reply.
I think the video series you shared is what I needed. Ill watch them and will try use embeddings.
Again, thank you very much for the help.

1 Like

Alex, did you try pushing the text in as-is form

I an new to AI. I want to start working with small example like take some text and then convert into Embedding vectors using any of the methods. Can you please help me if there are any available docs/videos for this to kick start.

Here you go:

Raymond Davey Embeddings

OpenAI Embedding Tutorial – BlinkData