Back in Oct 2019 Glow did not support ONNX models with Conv1D . A workaround was to use Conv2D instead. Now I see a recent merge for Conv1D (https://github.com/pytorch/glow/pull/4317). Does it mean there is a Conv1D support?
Yes, it should work!
@jfix, I can confirm the Conv1D in ONNX is supported. I have a follow up question, however.
Since Conv1D uses Conv2D by transposing/reshaping tensor, can you please confirm transposing and reshaping operations are cheap? I see lots of extra/new reshape and transpose nodes in graph.dot, whereas ONNX does not contain any.