Training with Mass pdf Data

Hello,

  1. I want to upload about 300 pdf documents to train my GPT about a specific topic, is there any way to simplify this task.
  2. If there is an update of a document how can i tell to GPT to not concider old versions of it.
    Thank you

20 is max for onboard doc load. But you can drag and drop or upload 10 files at a time into chat.

3 Likes

You just tap the corner of the pink doc icon in the gpt training window to delete ones you no longer need. Then upload the new version. The issue with using doc to train is it starts to confuse it with instruction/code in knowledge.

4 Likes

Hi, welcome.

CustomGPTs don’t work how you’re envisioning.

300 pdfs is a super-huge amount of information for a single cGPT to handle, not just considering the 20 file maximum.

For a cGPT you’ll want to specialize it in a given task or area of your PDFs. It can act more accurately from a spreadsheet wherever possible—if it’s possible to collapse your data from PDF to CSV, I’d look into it.

And you’re responsible for manually keeping the knowledge base clean and up to date. (The GPT’s don’t do that great when told to ignore sections of their live knowledge base. Just remove the files.)

If you really want to have that large of a Knowledge Base, look into Assistants.

2 Likes

Thank you
Unfortunately my data can not be collapsed to CSV format (Law texts, standards …)
I guess cGPT is not the solution for my project.
If i host a trained LLM based on GPT on a server, can i make cGPT call that external LLM ?