Brendan Folie

Thanks for asking @BAMcvoelker . To be honest we hadn't thought about it in a while, but after seeing your comment we realized we have all of the tools and...

Currently there is no method to sample from an RF distribution, though one could be easily created. A BoTorch [model](https://botorch.org/docs/models) only needs to produce a [posterior](https://botorch.org/docs/posteriors), which only needs to...

Hi Prateek. Lolo training is not entirely reproducible because the base learners are trained in parallel and we don't use splittable random numbers. This is a known deficiency, but I...

Sorry @ysiraichi, you're right that I neglected an important fact -- after hitting recompilation it hangs for an indeterminate period of time. I'll update the issue.

@bhavya01 I don't think that #9315 is related as it covers compilation of reduce_scatter. This is not a compilation issue. I believe it's a failure of `reduce_scatter_bucketized`, which exercises a...

One problem is [here](https://github.com/pytorch/xla/blob/master/torch_xla/csrc/cross_replica_reduces.cpp#L544), where it calls `ShapeHelper::ShapeOfXlaOp(reduce_result).dimensions_size() == 0`. Previously, if the result was a tuple of shapes, `.dimensions_size` would return 0. Now it crashes ([XLA PR here](https://github.com/openxla/xla/pull/24375)). The...

I don't think so. `.tuple_shapes()` will work as long as `all_gather_result` has a tuple shape, and I think in that context it does. But I haven't tested it

> For "Send and Recv", aren't the referenced passes for GPU XLA? I am not sure the motivation there was that Send/Recv was not suitable for XLA, but rather that...

> I believe "supporting native PT code that uses torch.dist.send/recv" is definitely a valid point, but slightly orthogonal. There may be some miscommunication, because that _is_ the point of this...

> IIRC, for [#8074](https://github.com/pytorch/xla/issues/8074), TRN actually worked fine, whereas GPU needed the "hidden attributes" - do you know if TPU has the same requirements? Yes, TPU requires the frontend attribute...