Custom GPT Model Training for Unversity LMS Courses (E-Learning)

I need to train the GPT model on my University courses for the students. The data is in the form of handouts, slides, videos, assignments, quizes etc. Please guide about the best practical successfully approach to acheive this task with minimum budget, reliable response and control.
I have explore these option but still confused about the right directions to acheive the ultimate goal.

  1. Create/Maintain/Build our own vector database (i.e. PineCone) of training material and use the open-source frameworks to push data (e.g., PDF/text) into the database and query the data using the OpenAI GPT model.
  2. Fine-tune OpenAI model, which is using JSONL data for training (OPenAI Own Vector Database Backend).
  3. Utilize the OpenAI Assistant API beta, which accepts files (OPenAI Own Vector Database Backend).

Please share your valuable suggestions.

You should be able to combine #2 and #3. Assistants use models. The model can be the base model or a fine-tuned model. When you fine tune a base model you get your personal model name that looks something like gpt-3.5-turbo-someRandomString. You can then pass it into your assistant and it’ll use it for better reasoning. As for the files, you’ve noted correctly that the assistant implements its own vector database, so you can just attach your files there.

1 Like