expose latencies with the speedup in OnnxDiscrepancyCheck

[xadupre]

Describe your changes

Speedup is not enough as a metric, latency must be added to understand which side is changing.

[xadupre]

@copilot extend the existing unit test to cover the changes

[Copilot]

@copilot extend the existing unit test to cover the changes

Done in 72c5d4e: I extended the existing discrepancy-check unit tests to cover the new latency-and-speedup return path from _measure_speedup.

[Copilot]

Pull request overview

Note

Copilot couldn't run its full agentic review because no GitHub Actions runner was available. Make sure your repository has a runner available to run Copilot's review, or add a copilot-setup-steps.yml file specifying one with the runs-on attribute. See the docs for more details.

This PR enhances OnnxDiscrepancyCheck speed measurements by exposing the underlying per-iteration latencies (PyTorch and ONNX) in addition to the speedup ratio, making it easier to interpret which side changed.

Changes:

• Updates _measure_speedup to return (pytorch_latency_s, onnx_latency_s, speedup) instead of only speedup.
• Stores pytorch_latency_s, onnx_latency_s, and speedup into the results dict when timing is measured.
• Adds a unit test verifying the new _measure_speedup return tuple.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
olive/passes/onnx/discrepancy_check.py	Returns and records both latencies alongside speedup in results.
test/passes/onnx/test_discrepancy_check.py	Adds coverage for the new tuple return contract of _measure_speedup.