[bug][AutoDeploy]: Phi-4 path fails with a shape mismatch error

System Info

• CPU architecture: x86_64
• GPU properties: not captured from CI artifact
• Libraries
  • TensorRT-LLM branch or tag: main (pipeline context)
  • Container used: AutoDeploy CI container
• OS: CI environment

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Affected registry entries currently disabled in examples/auto_deploy/model_registry/models.yaml:

• microsoft/phi-4
• microsoft/Phi-4-reasoning
• microsoft/Phi-4-reasoning-plus

All three fail in the AutoDeploy path with the same runtime shape-mismatch error.

Representative failure:

RuntimeError: a and b must have same reduction dim, but got [s44*s70, 5120] X [2560, 5120].

This looks like one shared Phi-4 family issue rather than three separate model-specific problems.

Expected behavior

Phi-4 entries should build a valid AutoDeploy graph and run without runtime shape mismatches.

actual behavior

The Phi-4 entries fail in AutoDeploy with a runtime shape-mismatch error before successful execution.

additional notes

NA

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

Start with the disabled entries in examples/auto_deploy/model_registry/models.yaml and reproduce the failure for microsoft/phi-4 or one of the related Phi-4 entries in the AutoDeploy CI container. Trace the reported reduction-dimension mismatch through the AutoDeploy path. Done means all three Phi-4 registry entries build a valid graph and execute without the runtime shape-mismatch error.