New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[rllib] Allow for JAX framework #8732
Comments
@sven1977 should we support framework=None or something like this? |
Yes, that would probably be more sustainable. Am I right in thinking that as long as my And if that's the case, why pick a default framework at all? |
Would RLlib allow for mixing of frameworks? What would go wrong if I use tensorflow to collect experience, pytorch to learn and jax/numpy to run trails? Of course, we can think of many reasons not to use such a setup, but to what extent does RLlib's functionality depend on a choice of framework? |
Yeah I think the framework parameter might have been a bit heavy handed. We do allow mixing of policies with different frameworks. I made a PR to remove the framework import checking, which should allow any kind of policy to be used no matter what the setting is. |
Yeah, let's allow |
@sven1977 This might be silly question, but why would we want to check the framework at all? I feel that with the latest changes that @ericl made in #8748 is a better setup, i.e. drop the framework checks altogether. This means that the framework config setting is just a hint that allows you to write some conditional logic if a specific value is set. For instance, I might want to implement some logic if |
Ok, this works now. Just explicitly use
Closing this issue. |
Good to hear that JAX is on your road map. Let me know if I can help. I shared a couple testing scripts in #8776 |
@sven1977 Is there a good issue to keep an eye on, RE: JAX support? Poking around in the |
Hi @sven and @dynamicwebpaige, #8776 is fixed, hopefully, by adding |
What is the problem?
I've been using JAX as my framework for a little while now. I just upgraded to the nightly build (due to some unrelated issues) and now RLlib is telling me I need to install TensorFlow or Torch.
I tried setting
{'framework': 'jax', ...}
in my trainer config, but this results in another error. Basically, not recognizing any framework other than one of:[tf|tfe|torch|auto]
Ray version: ray-0.9.0.dev0, Python 3.8, Ubuntu Linux 20.04 LTS
Script to reproduce:
If we cannot run your script, we cannot fix your issue.
The text was updated successfully, but these errors were encountered: