Increasing context length for language input

#39
by chrishoertnagl - opened

The model currently supports 4k context window, which for some mutli-turn conversation cases is a bit low. So I have 2 questions:

  1. Do you plan on increasing the context window in the near future
  2. How would you recommend (maybe some research or GH repos) to do it, if I have multi-turn text + image conversation data

Thanks :)

Hey @chrishoertnagl

  1. Yes, we plan to increase the context window in our upcoming release.
  2. Maybe implement memory mechanism that encodes past images and text responses separately, then feeds a summary or extracted features to the model.

Sign up or log in to comment