Eclectic-Sheep / sheeprl

Distributed Reinforcement Learning accelerated by Lightning Fabric
https://eclecticsheep.ai
Apache License 2.0
275 stars 26 forks source link

Fix/evaluate no grad #201

Closed belerico closed 5 months ago

belerico commented 5 months ago

Summary

This PR fixes a problem related to the increasing memory consuption for dreamer-v3 during the evaluation phase. In particular it let the user choose whether to disable gradients during the evaluation phase with the config parameter disable_grads, i.e. one can launch an evaluation with:

python sheeprl_eval.py checkpoint_path=/path/to/checkpoint.ckpt disable_grads=True

Type of Change

Please select the one relevant option below:

Checklist

Please confirm that the following tasks have been completed:

Thank you for your contribution! Once you have filled out this template, please ensure that you have assigned the appropriate reviewers and that all tests have passed.