jxtps

pytorch blackwell sm_100 support?

3

I'm trying to run on ubuntu + Nvidia B200, and it seems like libtorch_cuda.so bundled with the cuda presets does not support sm_100? ``` $ cuobjdump --dump-ptx libtorch_cuda.so ... Fatbin...

pytorch: HalfIndexer can't read/write short[]

3

The HalfIndexer can as of this writing only operate on doubles & floats, and there's a full complement of gets & puts to that end. However, in my application I...

Missing pytorch snapshots?

2

https://github.com/bytedeco/javacpp-presets/tree/master/pytorch references `2.8.0-1.5.13-SNAPSHOT` and `12.9-9.10-1.5.13-SNAPSHOT` but these appear to not exist on sonatype.org: ``` [error] (update) sbt.librarymanagement.ResolveException: Error downloading org.bytedeco:pytorch-platform-gpu:2.8.0-1.5.13-SNAPSHOT [error] Not found [error] Not found [error] not found: C:\Users\admin\.ivy2\localorg.bytedeco\pytorch-platform-gpu\2.8.0-1.5.13-SNAPSHOT\ivys\ivy.xml...

Reading https://github.com/google-research/nested-transformer/blob/main/models/nest_net.py#L107 and https://github.com/google-research/nested-transformer/blob/main/libml/self_attention.py#L266 there's a 3x3 convolution at each change in resolution (so 2 total). Did you get a chance to do any ablation studies on this? E.g. comparing...

Reading https://github.com/google-research/nested-transformer/blob/main/models/nest_net.py#L89 and https://github.com/google-research/nested-transformer/blob/main/libml/self_attention.py#L225 it's clear that the PositionEmbedding is done over *three* not two axes. This is further corroborated by the pytorch clone: https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/nest.py#L204 So it's taken over `(blocks,...