failed gemma3 convert checkpoint

[BayRanger]

python package info

tensorrt-llm 1.1.0rc0,: transformers==4.55.0

problem

Follwing the instrucrtions listed in models/core/gemma/README.md, the convert_checkpoint would fail with the error tensorrt_llm/layers/normalization.py", line 100, in postprocess return weights.to(getattr(torch, config.dtype)) ^^^^^^^^^^

Prolem

I doubt if currently the tensorrt no torch rontine really supports gemma3 inference. And how to fix it?

[BayRanger]

https://github.com/NVIDIA/TensorRT-LLM/pull/3247/files#diff-cf1241608c0319b85b9bc89542064a753024c19387bd2ee07abd5435e1ba5578 by implement this commit, the problem could be solved.