Can we add a performing inference example?

[bhack]

Currently we have a training example with "proxy-optimal" parameters in the repo.

Can we have an inference job example with "proxy-optimal" prams?

I think that an inference job example needs to have the prams for fast reading the input and fast saving the output (same or dedicated mounted buckets).

Generally inference is one time pass so many caching params are a bit useless to reach peak performance and it will be a good case to guide the user on the best practices.

[hime]

Hi @bhack, can you give us more details regarding "proxy-optimal" parameter and training example? I am not aware this example is present in the repository.

[bhack]

I meant this is "optimized" with params for training:
https://github.com/GoogleCloudPlatform/gcs-fuse-csi-driver/blob/main/examples/pytorch/train-job-pytorch.yaml

Can we have something similar for inference?

[hime]

We recently updated our documentation and have a optimized example for inferencing below. Hope this helps!

https://cloud.google.com/kubernetes-engine/docs/how-to/cloud-storage-fuse-csi-driver-perf#inference-serving-example

[bhack]

Ok