Does LlamaIndex consume ChatGPT tokens

depends on the index you use

vector index will call an embedding model during index construction

list index doesn't consume anything
tree index will call the LLM (which can be ChatGPT if you specify that)

Is LlamaIndex use openai gpt3 embedding api to generate embedding index?

yes by default

I am also working on a similar project and so far I can tell that the default settings doesn't quite work well for code files from my experience. You need to craft high quality prompt templates and find the best way to feed them to indices.

I found that just simply reading all the code files as a list of document (each code file is one document) and then feeding that list to a SimpleVectorIndex doesn't quite work as you would expected.

You definitely need to experiment with how to load code files into documents and best combination of using indices together.