How to teach your gpt to a new version of python library?

Hello, for the past few days, I have been trying to create a GPT that would understand the new version of the Aiogram 3 library in Python. I am facing an issue where it seems to refuse to accept the knowledge I have given it. The knowledge I provided is a 630-page PDF documentation, weighing 2.3 MB. Here’s what I have done:

  1. I gave it the documentation and told it to study it, referring to it whenever it wants to give an answer or explain something.
  2. I provided nearly 50 examples of different bot coding in the Create tab.
  3. I added to its knowledge base files with code examples in txt format.
  4. I experimented with writing instructions.

I would like to know about your experience. How did you resolve this, and did you resolve it at all? Did you face similar problems? What stage are you at now, and how do you actually teach GPT?

Thank you for your attention.

Hey there and welcome to the community!

I think there may be some misconceptions as to what “knowledge” files are and what you’re trying to do.

630 pages is way too much.

Anytime knowledge is retrieved, it is placed in its context window. So, by providing this massive amount of data, you’re essentially clogging your own context window. I would break things down a bit into smaller chunks (or GPTs).

1 Like

Thank you for the response and for providing at least some understanding of GPT. Could you tell me if there is any kind of manual or guide regarding this, if you know, please? Thank you for paying attention.

1 Like

Of course!

What in particular are you looking for in a guide or manual? There’s a lot to touch on here, and there’s no single consolidated guide or manual for both understanding how LLMs work, and how to effectively use and create custom GPTs. The latter comes easier after you understand the former. In most cases as well, adjusting and guiding is done on a case-by-case basis, because everyone’s needs are a bit different, and tend to require some different approaches. This is a very new and emerging field, so there’s no “catchall” manual yet.


If you haven’t yet, check out the Quickstart guide for some basics to get you started.

1 Like

Got it, thanks, I’ll experiment with gpt.