search

facebookresearch / Pearl

A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta.
MIT License
2.63k stars 159 forks source link

Model Export #45

Open wglassly opened 9 months ago

wglassly commented 9 months ago

Need Export API

rodrigodesalvobraz commented 9 months ago

Thanks, added to our TODO.

cryptexis commented 8 months ago

+1 to that.

The library is very powerful indeed. However, Pearl is about productionizing RL pipelines. And there is no information neither in documentation nor in form of a tutorial on how to export the agent/policy and how to use in production.

rodrigodesalvobraz commented 8 months ago

Thanks for the suggestion. We currently use a Meta-internal method for saving models which won't work externally. We are prioritizing addressing this issue in the open source version.

AliceHZhu commented 8 months ago

Is there an estimate on how soon exporting/loading trained agent/policy (better with checkpoints) will be available?

rodrigodesalvobraz commented 8 months ago

Is there an estimate on how soon exporting/loading trained agent/policy (better with checkpoints) will be available?

Thanks for checking. This is one of our highest priorities (along with others such as having more tutorials). I hope we can get this done in the next couple of weeks. I will post a comment here when it's done.