Hi OpenAI,
I created an application using GPT 3.5 Turbo API in CLI mode with parallel prompting. Any tips on parallel prompting?
Example:
prompt 1:
prompt 2:
prompt 3:
.
.
It will query asynchronous
What type of prompting is best for it for alpha testing purposes?
Well, these LLMs aren’t designed for “parallel prompting”, so no matter how you handle the prompts, at the end of the day, one thing’s gotta go in so one thing can go out.
This is very specific and dependent on how you want the prompts used as context for a response. You would need a data structure of your own design to handle this the way you want. If you’re trying to do something like real-time, multi-speaker input streaming, this is not possible through APIs like this, nor are most LLMs designed for this at this point in time. You would need to work with your own model from scratch, and completely gut the traditional interface.
Perhaps consider pooling several prompts at once, making it one big prompt. Or handle them one at a time. Or, maybe a queue-like data structure? It all depends on your needs and how you want to organize the data and input, and what parts would be relevant for a particular response. These are questions you would need to decide on.
Just note, you cannot escape turn-taking. Meaning, One query will always be given for one response to occur. Each speaker takes a turn speaking, one at a time.