I uploaded a 400 page file and it seems

what index are you using?

I just the Paul graham essay text with my own

Thanks for the reply - I uploaded all fed speeches from the past 2 years and it seems to only reply based on the first few speeches

so you're using the GPTSimpleVectorIndex. By default it just returns the most similar chunk (~4000 words). Try setting chunk_size_limit to a smaller number when creating the index e.g. index = GPTSimpleVectorIndex(..., chunk_size_limit=256) and setting similarity_top_k to a value > 1 during query: index.query(..., similarity_top_k=5, response_mode="compact")

Thanks. My outputs are a bit better but is there a way I can make them longer?

Or is it limited to 4k max token output?

you can make them longer (though don't think you can make outputs longer than 4000 tokens). keep in mind combined length (including prompt + output) is 4096 or so, you can adjust max_tokens in the custom llm's section here: https://gpt-index.readthedocs.io/en/latest/how_to/custom_llms.html

Got it thx