Skip to content

CVS-165537 Minor fixes to partially enable contrib op tests#1050

Merged
RajeevSekar merged 6 commits intoovep-developfrom
rajeev/contrib_ops_test
Apr 23, 2026
Merged

CVS-165537 Minor fixes to partially enable contrib op tests#1050
RajeevSekar merged 6 commits intoovep-developfrom
rajeev/contrib_ops_test

Conversation

@RajeevSekar
Copy link
Copy Markdown

Scope of the ticket was for the following 4 contrib ops:

onnxruntime/test/contrib_ops/embed_layer_norm_op_test.cc
onnxruntime/test/contrib_ops/matmul_4bits_test.cc
onnxruntime/test/contrib_ops/dynamic_quantize_matmul_test.cc
onnxruntime/test/contrib_ops/fused_matmul_op_test.cc

Fixes:

  1. dynamic_quantize: no change because absolute error between OV and expected is too high, so test remains skipped
  2. embed_layer: OpenVINO EP only supports 2 outputs for EmbedLayerNormalization, skip only when 3rd output (embedding_sum) is requested
  3. fused_matmul: Disable OpenVINO for scale (alpha != 1.0) and transBatch cases due to OV compilation errors., other than that run with OVEP.
  4. matmul_4bits: error thresholds added for float_32 cases and compile time skipping removed

Copy link
Copy Markdown

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR updates several contrib-op unit tests to run more broadly with the OpenVINO Execution Provider (where supported), while keeping known-incompatible scenarios excluded to avoid OpenVINO compilation/runtime failures.

Changes:

  • Enable MatMulNBits float32 test coverage with OpenVINO by adding explicit error tolerances and removing the USE_OPENVINO compile-time skip.
  • Update EmbedLayerNormalization tests to only exclude OpenVINO when the optional 3rd output (embedding_sum) is requested.
  • Refine FusedMatMul test execution provider exclusions so OpenVINO is only excluded for alpha != 1.0 and transBatch cases.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File Description
onnxruntime/test/contrib_ops/matmul_4bits_test.cc Adds default tolerances for float32 path and removes OpenVINO compile-time gating for float32 accuracy tests.
onnxruntime/test/contrib_ops/fused_matmul_op_test.cc Makes OpenVINO exclusion conditional (scale/transBatch only) while still excluding other unsupported EPs.
onnxruntime/test/contrib_ops/embed_layer_norm_op_test.cc Allows OpenVINO to run when only 2 outputs are used; excludes it when the 3rd output is requested.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread onnxruntime/test/contrib_ops/fused_matmul_op_test.cc
Comment thread onnxruntime/test/contrib_ops/fused_matmul_op_test.cc
Comment thread onnxruntime/test/contrib_ops/matmul_4bits_test.cc
Copy link
Copy Markdown

@preetha-intel preetha-intel left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@RajeevSekar RajeevSekar merged commit e662bf1 into ovep-develop Apr 23, 2026
7 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants