Hi all, is there any support for tensors with dynamic shape in TVM? For example, the shape of the output tensor is to be decided during execution according to the input data? Or is there any efforts now to support such features?
Not supported yet, and you maybe interested in looking at Does TVM support dynamic range sum? @jknight may have a better answer to this.
It is work in progress(in Nov’s Bay Area Meetup, there is a session: https://drive.google.com/file/d/1FaOQRN6fDna50l_iOjeOLeJHL7w_NFdz/view). @haichen @kevinthesun @yongwww @jroesch
1 Like
Thank you for your kind reply? So is it still in progress but not ready?
It seems that it is already work in progress, but not ready yet.