[BUG] Gemma3 Compatibility

[wns823]

Description

I encountered an error when using gemma3 models with the smolagents library.

• AttributeError: 'Gemma3Config' object has no attribute 'vocab_size'

The Issue

In src/smolagents/models.py,

try:
    self.model = AutoModelForCausalLM.from_pretrained(
        model_id,
        device_map=device_map,
        torch_dtype=torch_dtype,
        trust_remote_code=trust_remote_code,
    )
    self.tokenizer = AutoTokenizer.from_pretrained(model_id)
except ValueError as e:
    if "Unrecognized configuration class" in str(e):
        self.model = AutoModelForImageTextToText.from_pretrained(model_id, device_map=device_map)
        self.processor = AutoProcessor.from_pretrained(model_id)
        self._is_vlm = True
    else:
        raise e

I recommend adding an appropriate exception block, as shown in the example below.

except AttributeError as e:
    if "'Gemma3Config' object has no attribute 'vocab_size'" in str(e) :
        self.model = AutoModelForImageTextToText.from_pretrained(model_id, device_map=device_map)
        self.processor = AutoProcessor.from_pretrained(model_id)
        self._is_vlm = True

[aymeric-roucher]

Maybe the solution is just to switch line:

try:
    self.model = AutoModelForCausalLM.from_pretrained(

to:

try:
    self.model = AutoModel.from_pretrained(

Could you try that @wns823 ? And if it solves the issue, open a PR?

[Sakalya100]

Hi @aymeric-roucher

I have worked on this issue and tried the changes as suggested by you. This change works and it solves the issue.
Have opened a PR #1023 for the same

This is my first contribution to this repo and I have started contributing to open source recently. Please let me know if I am missing anything or something needs to be improved.

Thanks!