Brendan Folie

Splittable random numbers for reproducible training

3

[Bagger](https://github.com/CitrineInformatics/lolo/blob/main/src/main/scala/io/citrine/lolo/bags/Bagger.scala#L101-L110) and [MultiTaskBagger](https://github.com/CitrineInformatics/lolo/blob/main/src/main/scala/io/citrine/lolo/bags/MultiTaskBagger.scala#L51-L56) both train the individual models in parallel. Because the order of training is uncontrolled, this means that Lolo random forests are inherently non-reproducible, even if the bagging...

For Python >= 3.12, [test_nodes.py:test_eq](https://github.com/omry/omegaconf/blob/ceec6f45046d6d67b2e3faba32d8f2d3561ea81c/tests/test_nodes.py#L833) fails with parameters `node = PathNode(Path("hello.txt"))`, `value = "hello.txt"`. [Example on CircleCI](https://app.circleci.com/pipelines/github/omry/omegaconf/3091/workflows/b1f2e711-4bf5-48d1-86d1-37edd6791ba9/jobs/12199). This is because `Path(str)` and `str` have the same hash as of Python...

[RFC] Improved coverage for native distributed collective operations

7

## 🚀 Feature Improve coverage of the [PyTorch collective operations](https://docs.pytorch.org/docs/stable/distributed.html#synchronous-and-asynchronous-collective-operations) so that native distributed code is more likely to work without any user modification required. Collective ops would ideally work...

#9315

DDP does not work with torch_xla.compile

2

## 🐛 Bug Model wrapped by PyTorch's DDP can be compiled and run, but on the second step they hit an unexpected recompilation and hang indefinitely. This can be seen...