Upgrade transformers to 4.45.2

[cupertank]

I'd like to use vLLM together with TensorRT-LLM in my project, but now it's not possible because of transformers versions.

vLLM requires transformers >= 4.45.2 and TensorRT-LLM has upper bound <= 4.45.1.

Is it possible to upgrade a minor version of transformers in TensorRT-LLM?

[VALLIS-NERIA]

We have not tested yet but you can manually install a higher version of transformers. Usually there won't be much problem.

[cupertank]

@VALLIS-NERIA it's not possible to do correctly with pypoetry, we cannot have a consistent lock file. I think there won't be problems with upgrading transformers by one minor version for future TRT-LLM releases. We already tested it with trnasformers 4.45.2, and it's totally fine

[Xarbirus]

Hi @VALLIS-NERIA! I have the same problem, could you please give it more attention?

[VALLIS-NERIA]

Hi @cupertank @Xarbirus , you can edit the requirements.txt if needed. Upgrading transformers offically is not that trival because TRT-LLM has to support many models. Maybe you have already noticed some models in examples directory have different transformers version in their own requirements.txt.