add fast test to check with bfloat16

[xadupre]

fix discrepancy tests for bfloat16

[xadupre]

@copilot in test_bf16_precision, do you check that speedup is here with no error?

[Copilot]

@copilot in test_bf16_precision, do you check that speedup is here with no error?

It wasn't — fixed in the latest commit. The run call now passes --test_metrics mae,speedup to enable timing, and three assertions were added after it:

1. TEST_OUTPUT_MARKER_FILE exists in the output directory (confirms the run finished without error).
2. discrepancy_check_results.json exists.
3. The speedup key is present, is a number, and is > 0.

[Copilot]

Pull request overview

This PR adds a new CLI smoke test intended to exercise the olive optimize + olive run --test flow when targeting BF16 precision, and to validate that the discrepancy report is produced with a speedup metric.

Changes:

• Add test_bf16_precision / _assert_bf16_precision to run an optimize+run workflow with --precision bf16.
• Validate that discrepancy_check_results.json is written and includes a numeric, positive speedup value.

[xadupre]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Resolved in 3058fc9.

[xadupre]

@copilot it does not seem related to my changes test/passes/onnx/test_mnb_to_qdq.py:154: in test_mnb_to_qdq
with pytest.raises(Exception, match="uint8"):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E Failed: DID NOT RAISE Exception

[Copilot]

@copilot it does not seem related to my changes test/passes/onnx/test_mnb_to_qdq.py:154: in test_mnb_to_qdq
with pytest.raises(Exception, match="uint8"):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^...

I checked this branch against origin/main: the only PR diff is test/cli/test_cli_test_model_smoke.py, and the latest PR workflows for xadupre/bf16 are green. test/passes/onnx/test_mnb_to_qdq.py came in from the merged main side of 81ea757, so I did not make a code change for that failure.