joonspk-research/generative_agents

How to accelerate simulation time?

Opened this issue · 1 comments

I am trying to use this project. I found that although each step is 10 seconds, due to hardware or network latency, each step is actually greater than 10 seconds. If I want to observe a long stage, the time spent will be very long. Is there any way to speed up the simulation time?

I have looked at this improvement for more than a month, not lots of results tho :(

I was thinking to start looking to the architecture to understand if it is possible to parallelize some OpenAI calls. In the end I have never done it but my rough guess is that the main bottleneck are the model calls.