Investigate what work is required in systems side for serving TF session based models. What kind of SLA do we need for practical performance. Under what batch size/data.

[viswa-nvidia]

No description provided.

[viswa-nvidia]

@karlhigley , for completing the example in 22.10 , do we need to complete this investigation ? Are we already working on this ?

[karlhigley]

There are no models to investigate this with yet AFAIK, so I'd be very surprised if we completed an end-to-end example in 22.10

[karlhigley]

I also have no way to independently determine SLAs other than pulling a number out of the air, so the title of this issue doesn't make much sense 😅