Can dataflow compile code for GPU

Question

Since dataflow already creates a highly parallel framework, I was wondering if there was anyway to get our dataflow jobs to just "run" on google GPUs (and ideally a cluster of GPUs on different nodes) instead of a cluster of CPUs? It seems like it ought work in a similar manner to how tensorflow does it....

Answer 1

This feature is not supported at the moment, but we are considering it among other features for future prioritization.