big-AGI
big-AGI copied to clipboard
Gemini: auto-retry, auto-continue
Implement a "fixer" for Gemini, as the chance of calls breaking is very high, and token limits are often reached even earlier than expected. For this, implement:
- (easy) an auto-retry with exponential backoff up to 3 times for generation calls, which will retry on error and on the server side
- (hard) an auto-continue which can virtually extend the token size of the network, and will require a UI checkbox option for the user, so we can kick it in