masahi commented on a change in pull request #5383:
URL: https://github.com/apache/incubator-tvm/pull/5383#discussion_r412101164
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -337,6 +337,55 @@ def _impl(inputs, input_types):
return _op.transform.repeat(data, repeats=repeats, axis=axis)
return _impl
+
+def _parse_input_data(inp):
+ import torch
+ if isinstance(inp, _expr.Var):
+ data = inp
+ elif isinstance(inp, _expr.Call):
+ data = inp
+ elif isinstance(inp, torch.Tensor):
Review comment:
I still don't see how we get torch.Tensor as input? Inputs should always
be relay value. From Torch tensor, we should get the numpy array, and wrap
relay.const on it. Having torch tensor in op inputs is a logic bug.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
