Missing FIM Tokens?

#3
by dervagabund - opened

Hello,

I am just getting into running models locally to use with my IDE but I can't get the gemma 4 models to work with code completion.

Could it be that the fim tokens are missing in die tokenizer config?

Best regards,
Thomas

Google org
edited Apr 10

Hi @dervagabund
I attempted to reproduce the issue using the code completion example from the https://github.com/huggingface/huggingface-gemma-recipes/blob/main/notebooks/Gemma4_(E2B)-Multimodal.ipynb . It is working as expected . Could you please share the specific code snippet you are using? Will try to reproduce it on my side ? It would also be helpful to see any error logs or messages you're getting.
Thanks

the code completion in your example is create code from a screenshot. with code completion i am talking about is the fill in the middle logic for inline code completion when it's used in an ide.

I think that in general it works when the correct tokens are supplied, but the model does not report the tokens when loading it => the ide uses wrong tokens that are not recognized. i think the tokenizer_config.json is missing them.

hey Team
I am having error while trying E4B model:
OS Model benchmark aborted — google/gemma-4-e4b-it failed on 10 consecutive rows. Last error: 'list' object has no attribute 'keys'

Could you pls share the code snipped which automatically downloads the model from HF (i can put my creds of HF token) and uses the model with some text
I want E4B for text

Sign up or log in to comment