load a .h5 trained model directly from GCS ?

[YorelNation]

Hello, it's the fist time I actually try to put in a production environment a locally trained .h5 model. I have a website hosted on a cloud run container and I'm trying to run an Image processing pipeline every-time a file is uploaded to GCS via the website (that's why I want to use a cloud function that triggers when a new file is created).

my issue:

I have found a way to load my .h5 model from GCS but It's taking way too mush time and I'm sure there's surely a better way to do what i'm trying to do:

almost 1 minute to load on my local machine. Do you have any recommendation on how to trigger the prediction of my trained model + (pre/post processing) easily upon file upload from my website (in a serverless context) ?

[Horace]

Hello,

Can you kindly advise of your complete workflow. Additionally, how big are these h5 files?

Regards

[YorelNation]

I am using a could run docker container. And instead of packaging the model inside the container I would prefer to tell the code to load it directly from GCS (in python). Model .h5 size is 22.4Mo

[Horace]

Hello,

GCS does offer low latency which applies to all storage classes. If your app and GCS buckets are in the same region, I suggest to file an official support [1] so we can take a look at it in details using our internal tooling.

Regards,

[1] https://cloud.google.com/support-hub

[Horace]

Additionally, 

Have you tried comparing the results when fetching the same file directly through gsutil?