Issue with GPT4 in text transformation apps

I am using the GPT-4 model on several text transformation applications.

Recently, I have implemented an app that chains transformation tasks based on batches of documents.

See GitHub - mikiane/extended_llm: A simplified version of a second brain for neural language models using simple CSV files containing embeddings.

The app and the model work very well, but the performance of GPT-4 has been catastrophic compared to a few weeks ago. Response and generation times are extremely slow, to the point that these applications become almost unusable in streaming.

  • Is there a subscription to the API that could improve this situation?
  • I am almost forced to downgrade to version 3.5.
  • I am using the GPT-4 model and considering limiting myself to the version from last March to see if it makes a difference (but without benefiting from the latest updates).