Viktor Gyenes

@nvpohanh this is the issue I mentioned to you.

@guschmue yes I know it's for the batch dim, and I did check out the subgraph, that's okay. However, the model has dynamic input shapes, and if you fix those...

Indeed, it would be nice if that loop would disappear when batch size is 1, but I agree that it's probably hard to optimize it away completely. I could give...

The problem is not the time it takes to run the loop, but that our inference engine does not handle loops. The idea was to edit the graph and replace...

@guschmue any feedback on this?

I believe I have figured out what the problem is with this model. The model itself may apparently be valid, but it has a special case for the Loop operator...

Yes, there is a plan to provide a pip installer for NNEF-Tools. By the way, you are saying you run `python setup.py install` from the parser folder, whereas you should...

No exact deadline, but we hope to roll it out in the coming month at some point. Although the plan was to put out a source version that compiles itself...

@botev just to clarify if I get it right: are you saying that the structure description could be generic, not exactly specifying the shape of variables, and the shape would...

Okay, I understand what you mean now. I played around a bit with a similar idea, but thought it would be too complicated to implement (I did not think it...