Accessing the context which is passed to the LLM before a response is generated

[MortazaG]

Is there any way of getting access to the context which is passed to the LLM (eg. gpt-3.5-turbo) right before it is passed to the actual LLM for generating a response based on the context and user input?

I want to access the generated context from a RAG or semantic search, because I am curious about how it looks and also because I want to be able to more accurately calculate the amount of tokens used for each ChatCompletion when using OpenAI.

Is there any way of achieving this when using LlamaIndex?