Attempting to use Vertex Custom Training with GPU enabled, and I’m finding some oddities with access to GPU resources. I’m hoping someone has insights into what I might be doing incorrectly.
Below is the caller of the training code, with a chunk of the configuration, as well as the error message at the bottom - this docker image reportedly does not support GPU resources. Everything should be configured correctly, the worker pool is using highmemory-16, which supports gpus, and the sklearn GPU 1.6 image. What could be causing this issue?
However, this image is named sklearn-gpu, and supports GPU on this chart:
and lastly, the configured pool is using high-mem 16, which supports 1,2,4 accellorators of the given type.
