jwfromm commented on a change in pull request #4271: [Relay][Frontend][ONNX]
operator support: DepthToSpace, SpaceToDepth
URL: https://github.com/apache/incubator-tvm/pull/4271#discussion_r344929252
##########
File path: tests/python/frontend/onnx/test_forward.py
##########
@@ -77,22 +77,27 @@ def get_tvm_output(graph_def, input_data, target, ctx,
output_shape=None, output
return tvm_output.asnumpy()
-def get_caffe2_output(model, x, dtype='float32'):
- import caffe2.python.onnx.backend
- prepared_backend = caffe2.python.onnx.backend.prepare(model)
- W = {model.graph.input[0].name: x.astype(dtype)}
- c2_out = prepared_backend.run(W)[0]
- return c2_out
+def get_onnxruntime_output(model, x, dtype='float32'):
+ import onnxruntime.backend
+ rep = onnxruntime.backend.prepare(model, 'CPU')
+ x = x.astype(dtype)
+ ort_out = rep.run(x)[0]
+ return ort_out
def verify_onnx_forward_impl(graph_file, data_shape, out_shape):
dtype = 'float32'
x = np.random.uniform(size=data_shape)
model = onnx.load_model(graph_file)
- c2_out = get_caffe2_output(model, x, dtype)
- for target, ctx in ctx_list():
- tvm_out = get_tvm_output(model, x, target, ctx, out_shape, dtype)
- tvm.testing.assert_allclose(c2_out, tvm_out, rtol=1e-5, atol=1e-5)
+ try:
+ c2_out = get_onnxruntime_output(model, x, dtype)
+ except onnx.onnx_cpp2py_export.checker.ValidationError as e:
Review comment:
Oh interesting. The checker must be using the OpSet-1 version of
DepthToSpace, which doesn't have a mode argument. You should try updating your
version of onnxruntime to 1.0.0 (the same that tvm will use after the CI
update).
Alternatively for our test we could just use the default mode argument (i.e.
not specifying it). I think that would be better than having a try except. We
can add testing for mode some time in the future when we update to a newer onnx
version.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
