Alex Morehead

Hi, @jermainewang. Thanks for your quick response. This sounds like a good solution for use cases where we are only constructing our input DGLGraphs once, before the forward pass of...

Hi, @jermainewang and @lygztq. Thank you very much for your ideas here. To the best of my knowledge, the `knn_graph()` and `segmented_knn_graph()` functions both return combined graphs without batch metadata...

@LarsHill, I agree that this is an important question to ask. I am also encountering a use case where I need to track my custom `Metric`'s state with a dictionary....

Out of curiosity, has this build script successfully been converted into a feedstock for easy ingestion on PowerPC platforms like Summit?

Hi, @alexlang74. I would say yes, although at the moment I have found a few workarounds to get DeepSpeed set up on Summit. Having a feedstock, though, I can imagine...

For those interested in a quick solution until someone implements a native PyTorch approach, the following custom `Metric` may suffice. ```py import scipy import torch from torchmetrics import Metric class...

Hi, @stancld. My availability is quite limited at the moment. Are you referring to a full PyTorch-tensors implementation of this metric, or instead one that transfers label and prediction tensors...

@stancld, will do! I greatly appreciate your time and help here, and I'm sure others will as well!

@Miffyli @Merenion I would also be interested to see how @trigaten and @brandonhoughton have installed `xvfb` and configured MineRL to run on their copies of macOS (e.g., macOS Monterrey). For...

I second this original question. I think having this functionality working within PyTorch Lightning could greatly increase the visibility of this awesome repository!