Amog Kamsetty
Amog Kamsetty
Is there any plan to support user-defined procedures and the CALL interface?
Is support for PyTorch 1.9 on the roadmap for `mlagents`? Currently `torch` is constrained to be less than 1.9 in `setup.py`.
Warning messages (`warnings.warn`) are used in the `_Trainable`, but these don't actually get shown on stdout because they are inside an actor. I think the fix is to follow this...
Currently we directly raise any errors from tune or sklearn. We should add better error messages so the user knows how to resolve these errors. There are two types of...
`xgboost-ray` training currently does not work on Windows. Please thumbs up this issue or leave a comment if you would like to see Windows support!
As per @krfricke, we should create a single `RayCommunication` actor class that holds both an `asyncio Queue` and `asyncio Event`. This way we only need one remote actor for all...
Why would one want to have more than 1 cpu or gpu per actor? Is each actor parallelizing across its Cpus? What would be the benefits of adding more cpus...
Signed-off-by: Amog Kamsetty `ray.get_gpu_ids()` sometimes returns a list of ints or sometimes returns a list of strings depending on if the user has set the `CUDA_VISIBLE_DEVICES` environment variable. This has...
In the effort to have more realistic examples for Ray Train, we should port over the `timm` example (like what we had for Ray SGD v1 https://github.com/ray-project/ray/tree/master/python/ray/util/sgd/torch/examples/image_models) and also show...
Previously, `ray.get_gpu_ids()` would sometimes return a List of ints and sometimes return a List of strings depending on if the user has set the `CUDA_VISIBLE_DEVICES` env var, which made the...