delldu commented on a change in pull request #8443:
URL: https://github.com/apache/tvm/pull/8443#discussion_r670276742
##########
File path: python/tvm/relay/frontend/pytorch.py
##########
@@ -1875,7 +1875,9 @@ def Bool(self, inputs, input_types):
def Float(self, inputs, input_types):
assert len(inputs) == 1
- return _op.cast(inputs[0], "float32")
+ if isinstance(inputs[0], _expr.Expr):
+ return inputs[0]
+ return float(inputs[0])
Review comment:
> @delldu Please drop support for script mode. Just because your model
has control flow doesn't mean these ops also need to be scripted. Remember,
everyone can use these op conversion. You can do trace and selectively script
certain parts of your model using `torch.jit._script_if_tracing`. This is how
MaskRCNN in torchvision is implemented, for example. See
https://github.com/pytorch/vision/blob/master/torchvision/models/detection/roi_heads.py#L454
Thanks a lot. It is good reference.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
