Whether TVM will support dynamic shapes in the future

Development
1

I would like to know whether dynamic shapes are already supported in a planned way. I am not familiar with other deep learning frameworks, but without dynamic shapes, many OP of TF cannot be added, which leads to TVM’s narrow choice of models and lack of generality.
I seem to see some dynamic information in the recent PR, so if there is a plan, could anybodyplease tell me the approximate time?Thanks a lot for this.

2 Likes
2

Can anyone answer that? Thanks very much.

3

I think @yongwww mentioned in a previous posts that some people is working on this?. Maybe he can tell how far is this support.

In addition, there are multiple posts in this topic. I also hope to that TVM supports this in the near future

4

I believe @kevinthesun and @haichen are working on that.

5

We will have a RFC regarding dynamic shape support soon. To achieve descent performance, there is still lots of work remaining. A rough timeline for performant dynamic shape model can be late this year or early next year.

2 Likes
6

Thank you very much for your efforts in this regard.and im looking forward to it. Thanks again！

7

@kevinthesun is there any update on this work?

8

An RFC for the runtime dispatching has been filed here.

9

Hi I was wondering about the current status of this?

10

@kevinthesun @haichen any update on this work? I’m also quite interested in this feature.