Artur Niederfahrenhorst
Artur Niederfahrenhorst
Hi @rajveer43 , hi @tarunsamanta2k20 ! Thanks for offering your help! So, I think what you need to work on this is https://docs.ray.io/en/latest/ray-contribute/development.html#building-the-docs (to build the docs and see the...
@rajveer43 Thanks! In order to properly develop on Ray (excluding Ray Core parts), please follow the instructions at https://docs.ray.io/en/latest/ray-contribute/development.html#building-ray-python-only You should fork the repo and checkout master. If you are...
From there, you can start iterating on your problem by making changes and building the docs to see the effects.
Please interact with @justinvyu if you have any more questions in the coming days, as I will be on vacation.
Related [tensorboard](https://tensorboard.dev/experiment/JxgoXnfBQnexvYCjlRPTYw/#scalars&_smoothingWeight=0&runSelectionState=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%3D) that shows speedups on rollout worker side:
@kouroshHakha I've also added a configuration enumerator instead of relying on two long strings "complete_update" and "forward_train".
I looked into the logs and found that this is because we attemt to use the RL Module API in this test, which is, today, not ready. Same goes for...
I've included an example that prooves learning with a custom tokenizer. This custom tokenizer makes it so that the model is the same as in our custom_rnn example from the...