Running Gemma-2b with Torch 2.0.1?

#28
by insdaguirre - opened

Hi,

I am trying to run Gemma-2b-it using a server with 4 x Nvidia Titan XP's. However, this means that I am limited to Cuda 11.4, Torch 2.0.1, and Transformers 4.28.1. As such, I currently get this error:

"ValueError: Tokenizer class GemmaTokenizer does not exist or is not currently imported"

This is logical, as my Torch and Transformers are outdated. Since I cannot upgrade my torch further, are there any alternative ways to run Gemma-2b with this hardware? Using an API is not feasible because I will have an extremely high call count. Thanks!

Google org
edited Aug 21

Hi @insdaguirre , Could you please try running the model code again with the latest Transformers and torch version which you can update using the code - !pip install -U torch and !pip install -U transformers. Let us know if this works!. Thank you.

Sign up or log in to comment