masahi commented on a change in pull request #7747:
URL: https://github.com/apache/tvm/pull/7747#discussion_r603599665
##########
File path: tests/python/frontend/onnx/test_forward.py
##########
@@ -4278,6 +4278,33 @@ def test_wrong_input():
relay.frontend.from_onnx(model, shape=wrong_shape_dict)
+def test_aten():
+ torch.set_grad_enabled(False)
+
+ def _convert_to_onnx(model, inputs):
+ file_name = "{}.onnx".format("aten_model")
+ torch.onnx.export(
+ model, inputs, file_name, export_params=True, verbose=False,
opset_version=10
+ )
Review comment:
> Actually it looks like the version of torch we're using for CI doesnt
yet support `aten`. For now we'll have to add this without `aten=True` but we
can enable it later if we update versions. The good news is old versions of
torch always use ATen fallback ops so we'll be testing embeddingbag.
Really? The link I posted above is for PT v1.7 and this is what we run on CI.
--
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]
