vvchernov commented on code in PR #13866:
URL: https://github.com/apache/tvm/pull/13866#discussion_r1091526220
##########
tests/python/frontend/onnx/test_forward.py:
##########
@@ -7821,6 +7821,38 @@ def verify_sequence_ops(tensor_shape, num_tensors,
axis=0, position=0, new_axis=
verify_sequence_ops((3, 3, 3, 3), 4, axis=2, new_axis=1)
[email protected]_targets
+def test_empty_sequence(target, dev):
+ """test_empty_sequence"""
+
+ # Test creating an empty tensor sequence.
+ empty_node = helper.make_node(
+ "SequenceEmpty",
+ inputs=[],
+ outputs=["empty_sequence"],
+ )
+
+ length_node = helper.make_node("SequenceLength",
inputs=["empty_sequence"], outputs=["output"])
+
+ graph_outputs = [helper.make_tensor_value_info("output",
TensorProto.INT64, [])]
+
+ graph_nodes = [empty_node, length_node]
+
+ graph = helper.make_graph(
+ graph_nodes,
+ "Sequence_test",
+ inputs=[],
+ outputs=graph_outputs,
+ )
+
+ model = helper.make_model(
+ graph,
+ producer_name="Sequence_empty_test",
+ )
+
+ verify_with_ort_with_inputs(model, [], target=target, dev=dev)
Review Comment:
It is not so due to I use 2 nodes (SequenceEmpty and SequenceLength) in test
graph and the output is not empty, it is scalar with sequence length, as result
we compare zeros from TVM and ONNX Runtime output.
I do not think that default type influences on anything in this test
--
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]
