Only one program will be evaluated at the same time?

[linhaowei1]

I noticed that the distributed parameter in OpenEvolve appears to be unused:
https://github.com/codelion/openevolve/blob/bfb1c9d295ad1d36b24d2ff152494cd8a9918931/configs/default_config.yaml#L104C3-L104C14

Do you have plans to implement this feature? I looked into the run() function of the controller, and it seems that only one program is evaluated at a time.

I am trying a use case that evaluation is quite time-consuming, so I’m hoping to generate multiple child programs per iteration and evaluate them in parallel, and I cannot find this implementation. Do I miss something or it's just not implemented?

[linhaowei1]

I noticed that there is a evaluate_multiple() function in evaluator but not used. Can we apply it in controller directly?

[codelion]

I noticed that there is a evaluate_multiple() function in evaluator but not used. Can we apply it in controller directly?

Yes you can use it, it will evaluate on the same machine but in parallel using task pool.

Distributed processing is on the roadmap but not implemented yet. I was thinking actually something simple like e2b that allows running the eval on a remote sandbox can be used instead of trying to build the infrastructure for distributed processing. The LLM calls are all abstract with the APIs and if we can abstract the evals also with an API then we can scale easily. Please feel free to chime in and see if you want to implement a PR.

Thanks.

[codelion]

Fixed in #120