4697

[micwill755]

Description

Fixes test_base_fp4_static_shapes so that it compares equivalent quantized executions.

Previously, the eager reference output was generated before mtq.quantize() was called. As a result, the test compared an FP16 eager result against an NVFP4 TensorRT result. The expected quantization loss between FP16 and NVFP4 caused the numerical check to be flaky across GPU architectures.

This change moves the eager-model invocation after ModelOpt quantization. The test now compares the ModelOpt NVFP4/QDQ reference output against the TensorRT NVFP4 output.

This is consistent with the existing test_base_fp4_dynamic_shapes and test_nvfp4_nd_input_quantizer tests, which also generate their expected outputs after quantization.

Fixes #4697

Type of change

• Bug fix (non-breaking change which fixes an issue)

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my code
• I have commented my code where necessary
• I have verified the existing unit test with my changes
• New and existing relevant unit tests pass
• I have added the relevant PR labels