The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

Thanks! I'm going to hold on this for a moment - have to check and see if a nightly torch cu130 build is ready to fully test this. We also need to fix the backwards compatibility issues with CUDA < 12.9, since the included CCCL for those versions looks to be too old for using e.g. cuda::maximum. I suppose the options there will be to either guard around this or to pull in the newest CCCL separately.

Thanks! I'm going to hold on this for a moment - have to check and see if a nightly torch cu130 build is ready to fully test this. We also need to fix the backwards compatibility issues with CUDA < 12.9, since the included CCCL for those versions looks to be too old for using e.g. cuda::maximum. I suppose the options there will be to either guard around this or to pull in the newest CCCL separately.

for sure, really appreciate it, it is was only the start way. Feel free to change it

Thanks, I'm going to close this in favor of #1761 which takes this further to handle backwards compatibility, packaging, docs, etc.