get_model_config raises ValueError when using official variant 4b-it

[h-suzuki-isp]

The official example code uses the 4b-it variant with get_model_config, but this results in a ValueError. It appears that "4b-it" is not listed as a supported variant inside get_model_config, even though it is referenced in the official code.

This results in the following error:

ValueError: Invalid variant 4b-it. Supported variants are "1b", "2b", "2b-v2", "4b",, "7b", "9b" "12b", "27b", and "27b_v3".

Even if I change the variant to one of the supported variants, it causes errors in other parts of the code.

[Balakrishna-Chennamsetti]

Hi @h-suzuki-isp ,

Thank you for making us aware. I have reproduced the issue in my local, seems like the config file logic needs to be modified to handle the 4b-it variant.

I’d like to follow up with a few additional questions:
What exactly are you facing issues with supported variants.?
Were you facing any issues with executing the code, downloading model, loading, or anything else.?

Thanks.

[h-suzuki-isp]

Hi, @Balakrishna-Chennamsetti

Thank you for your response and for escalating the issue.

Following the error message, I attempted to switch the variant to "4b" instead of "4b-it". However, I encountered two problems:

• 1. The model weights failed to download.
• 2. When trying to instantiate the model, I received the following error:

ValueError: Unknown architecture: Architecture.GEMMA_3

This occurred during the initialization of the Gemma3ForMultimodalLM class. Here's the relevant snippet from the stack trace:

File /workspace/pytorch_gemma3/gemma_pytorch/gemma/model.py:485, in GemmaModel.__init__(self, config)
    raise ValueError(f'Unknown architecture: {config.architecture}')

This suggests that the architecture specified in the configuration (GEMMA_3) is not supported or recognized by the current implementation.

That said, I want to clarify that I haven’t deeply investigated the implementation, and it's entirely possible that I’m using it incorrectly or missing some setup step. So this might not be the root cause of the issue. Still, I hope this information helps in narrowing down the problem.

Thanks again for looking into this.

[Balakrishna-Chennamsetti]

Hi @h-suzuki-isp ,

The above mentioned is got resolved after merging the PR please check. Thank you so much for your patience.

Thanks.