Artur Niederfahrenhorst

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? We are still using our own deprecated API for caoncating samples. ## Checks - [x] I've signed off every commit(by using...

## Why are these changes needed? As discussed via Slack, we need to match the future ModelCatalog, the Models and RLModule in the sense of where to define specs and...

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? The replay buffer demo likes to take approx 10 iterations to reach the nice reward of 50 but often gets cancelled...

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? We currently modify the original sample batch in place when counting instead of the copy. (We make everything except SampleBatch.SEQ_LENS an...

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? Replaces imports of pygame with try_imports, akin to the ones we use for frameworks. This helps us output a more informative...

[RLlib] Terminated/Truncated in Quickstart Script

1

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? Our quickstart example in the docs has been updated to fit gymnasium, but a docstring is missing the truncated flag and...

[RLlib] Enhance input readers to handle complex combinations of trained policies vs offline data

1

Signed-off-by: Artur Niederfahrenhorst ## Why are these changes needed? Our input readers today simply throw an error if we want to ingest single agent data in a multi-policy training workflow...

[Tune] Some examples pages are not structured well

4

### Description Some of our Tune examples look a little weird ([example](https://docs.ray.io/en/master/tune/examples/lightgbm_example.html)). There is an index at the top with only one entry “Example”. Also, we dump some output that...

[RLlib] Compile update logic on learner and use cudagraphs

1

## Why are these changes needed? In the first attempt to leverage torch compile, we didn't introduce a compiled update method on the side of the learner (1) and also...

[RLlib] RNNs and RLModules

3

## Why are these changes needed? This is a fork of https://github.com/ray-project/ray/pull/30997 to experiment with how we incorporate RNNs into RLModules.