main bison
We managed to 2x our Azure GPT-4 rate limits, by using a simple model switching strategy
Decided to open-source it for everybody else - (feel free to DM me if you need help 🙂 )
Here's a tutorial: https://github.com/BerriAI/reliableGPT/blob/main/examples/Azure_OpenAI_Model_Switching_demo.ipynb
river root
in a nutshell if the azure limit is exceeded it will default to openais API (aka their models?) is that correct?
Other than the data security benefit, why would one choose to use the azure openai model over openai's hosted model. since one is rate limited while the other is not?
main bison
improved reliability