tqchen commented on code in PR #19863:
URL: https://github.com/apache/tvm/pull/19863#discussion_r3451531886
##########
tests/python/relax/test_frontend_onnx.py:
##########
@@ -5579,6 +5579,32 @@ def test_affine_grid(align_corners):
check_correctness(model, opset=20)
+def test_affine_grid_3d():
+ affine_grid_node = helper.make_node(
+ "AffineGrid",
+ ["theta", "size"],
+ ["grid"],
+ align_corners=1,
+ )
+
+ graph = helper.make_graph(
+ [affine_grid_node],
+ "affine_grid_3d_test",
+ inputs=[
+ helper.make_tensor_value_info("theta", TensorProto.FLOAT, [2, 3,
4]),
+ ],
+ initializer=[
+ helper.make_tensor("size", TensorProto.INT64, [5], [2, 3, 8, 16,
16]),
+ ],
+ outputs=[
+ helper.make_tensor_value_info("grid", TensorProto.FLOAT, [2, 8,
16, 16, 3]),
+ ],
+ )
+
+ model = helper.make_model(graph, producer_name="affine_grid_3d_test")
+ check_correctness(model, opset=20)
Review Comment:
seems we have been bringing in some of the tests that involves
check_correctness against onxx runtime.
I know that we have been doing so in past.
Unfortunately keep doing so would slow down our ci and and a better way
would be not doing integration tests in frontends and instead rely on cheaper
structural verify, would be great if we can hold off merging changes that
involves integration tests in frontends. For some of the key relax op), they
should be constructed as relax ast and run verification per OP as UT (perhaps
against pytorch.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
