Skip to content

Conversation

praasz
Copy link
Contributor

@praasz praasz commented Dec 1, 2022

Description

The test models in backend tests:

  • test_castlike_BFLOAT16_to_FLOAT
  • test_castlike_FLOAT_to_BFFLOAT16
    have changed between versions 1.11 and 1.12. The input 'Like' has got shape 3x4 instead 1 but data for this input is still shape of 1.

This change add consistency between model and data.

Motivation and Context

The OpenVINO project use this tests. After upgrade from 1.11 -> 1.12 these test start failing as there is no consistency between input data and inputs shapes.

@praasz praasz requested a review from a team as a code owner December 1, 2022 15:00
Signed-off-by: Raasz, Pawel <pawel.raasz@intel.com>
@praasz praasz force-pushed the bugfix/castlike-tests-model-like-data-shape-mistmatch branch from 0d6ccac to f77f056 Compare December 1, 2022 15:03
@daquexian
Copy link
Member

daquexian commented Jan 25, 2023

Thanks for your contribution!

However, it seems that the shape of like should be exactly [1] instead of [3, 4]

like = output.flatten()[0:1]

And I think the mismatch you encountered comes from that we specified a wrong type proto here:

input_type_protos=[input_type_proto, output_type_proto],

output_type_proto (whose shape is [3,4]) is incorrectly used as the type proto of like (whose shape is [1])

@praasz praasz force-pushed the bugfix/castlike-tests-model-like-data-shape-mistmatch branch from d0e9801 to bc1812e Compare April 25, 2023 13:35
praasz added 3 commits April 25, 2023 15:44
Signed-off-by: Raasz, Pawel <pawel.raasz@intel.com>
Signed-off-by: Raasz, Pawel <pawel.raasz@intel.com>
@praasz
Copy link
Contributor Author

praasz commented Apr 26, 2023

Thanks for your contribution!

However, it seems that the shape of like should be exactly [1] instead of [3, 4]

like = output.flatten()[0:1]

And I think the mismatch you encountered comes from that we specified a wrong type proto here:

input_type_protos=[input_type_proto, output_type_proto],

output_type_proto (whose shape is [3,4]) is incorrectly used as the type proto of like (whose shape is [1])

You've got right about proto. I've created like proto which use correct shape and generated test model will have correct inputs aligned with test data.

@p-wysocki
Copy link
Contributor

@gramalingam @jcwchen could you please approve necessary CI workflows?

@jcwchen jcwchen added the run release CIs Use this label to trigger release tests in CI label May 10, 2023
@jcwchen jcwchen enabled auto-merge (squash) May 11, 2023 04:30
@jcwchen jcwchen merged commit 745b9be into onnx:main May 11, 2023
@praasz praasz deleted the bugfix/castlike-tests-model-like-data-shape-mistmatch branch May 12, 2023 05:47
adityagoel4512 pushed a commit to adityagoel4512/onnx that referenced this pull request Jul 28, 2023
Signed-off-by: Raasz, Pawel <pawel.raasz@intel.com>
Co-authored-by: G. Ramalingam <grama@microsoft.com>
Co-authored-by: Chun-Wei Chen <jacky82226@gmail.com>
Signed-off-by: Aditya Goel <agoel4512@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
run release CIs Use this label to trigger release tests in CI topic: test
Projects
Status: Done
Development

Successfully merging this pull request may close these issues.

5 participants