Computing performance improvement.

[phalexo]

Since the "accelerate" module is used, I am able to put different stages on different GPUs (I have 4) and therefore avoid the RAM limitation issue.

But, this does not help with performance since different stages are run sequentially.

The question I have is whether it is possible to improve performance?

The immediate thing that comes to mind, maybe I can run the text model in parallel with the first/second/third stages?

[phalexo]

Also the model loading process into different GPUs can go on, more or less, in parallel. Suggestions on how to go about it, would also be welcome.