google

/

gemma-3-1b-it

like 199

Follow

Google 10.1k

pip install git+https://github.com/huggingface/transformers

    model_id = "google/gemma-3-1b-it"

    tokenizer = AutoTokenizer.from_pretrained(model_id)

    pipe = pipeline("text-generation", model=model_id, device="cuda", torch_dtype=torch.bfloat16, tokenizer=tokenizer)

from transformers import pipeline, AutoTokenizer

model_id = "google/gemma-3-1b-it"
tokenizer = AutoTokenizer.from_pretrained(model_id)
pipe = pipeline("text-generation", model=model_id, device="cuda", torch_dtype=torch.bfloat16, tokenizer=tokenizer)

KeyError: <class 'transformers.models.gemma3.configuration_gemma3.Gemma3TextConfig'>
File , line 4
1 from transformers import pipeline, AutoTokenizer
2 model_id = "google/gemma-3-1b-it"
----> 4 tokenizer = AutoTokenizer.from_pretrained(model_id)
6 pipe = pipeline("text-generation", model=model_id, device="cuda", torch_dtype=torch.bfloat16, tokenizer=tokenizer, token='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')
9 messages = [
10 [
11 {
(...)
19 ],
20 ]
File /local_disk0/.ephemeral_nfs/envs/pythonEnv-8256526d-05d2-4a9a-a39f-c9fefa37f7df/lib/python3.12/site-packages/transformers/models/auto/tokenization_auto.py:975, in AutoTokenizer.from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs)
973 model_type = config_class_to_model_type(type(config).name)
974 if model_type is not None:
--> 975 tokenizer_class_py, tokenizer_class_fast = TOKENIZER_MAPPING[type(config)]
977 if tokenizer_class_fast and (use_fast or tokenizer_class_py is None):
978 return tokenizer_class_fast.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
File /local_disk0/.ephemeral_nfs/envs/pythonEnv-8256526d-05d2-4a9a-a39f-c9fefa37f7df/lib/python3.12/site-packages/transformers/models/auto/auto_factory.py:771, in _LazyAutoMapping.getitem(self, key)
769 model_name = self._model_mapping[mtype]
770 return self._load_attr_from_module(mtype, model_name)
--> 771 raise KeyError(key)

what version of transformers you have? mine says 4.50.0.dev0, I installed today:

pip install git+https://github.com/huggingface/transformers

Same, I've tried it with both
pip install git+https://github.com/huggingface/transformers
and
pip install git+https://github.com/huggingface/[email protected]
:

"Successfully installed transformers-4.50.0.dev0"

Hm, strange, this is my code I ran today localy, not much but hope it helps, win11, python 3.12:

from transformers import pipeline, AutoTokenizer
import torch

class Helper:
def __init__(self, system_prompt):
    self.system_prompt = system_prompt

    model_id = "google/gemma-3-1b-it"

    tokenizer = AutoTokenizer.from_pretrained(model_id)

    self.pipe = pipeline("text-generation", model=model_id, device="cuda", torch_dtype=torch.bfloat16, tokenizer=tokenizer)


def generate(self, prompt):
    messages = [
        [
            {
                "role": "system",
                "content": [{"type": "text", "text": self.system_prompt}, ]
            },
            {
                "role": "user",
                "content": [{"type": "text", "text": prompt}, ]
            },
        ],
    ]

    output = self.pipe(messages, max_new_tokens=512)[0]

    return output[0]['generated_text'][-1]['content']

Same issue here. Any update?

pip uninstall transformers

pip install git+https://github.com/huggingface/transformers

works for me. it installed [email protected]

It seems like this might be a Python version issue.
I've tried using Python 2.13 along with all the variations here, but nothing seems to work.

Same error here is there any solutions? I'm using python 3.12
update: Official code snippet is broken use this instead
https://huggingface.co/google/gemma-3-1b-it/discussions/9#67d2f4acf1fd2d031480e92e

Also getting the same issue with Python 3.12

Edit: As mentioned above, the code snippet is broken. Use the example provided earlier within this thread, calling the class with system and user prompt.

You could try this demo on the ModelCard page.

You could try this demo on the ModelCard page.

Hey all! We've updated the transformers code to fix this issue.
I recommend reinstalling from the gemma3 release (or from main).

This is how you would do it:

Start by uninstalling transformers

pip uninstall transformers

Install from the Gemma-3 tag or main:

pip install git+https://github.com/huggingface/[email protected]
# Or if you want all latest changes, install from main
pip install git+https://github.com/huggingface/transformers

Hi @steve122192 ,

Tested the code with both the Gemma-3 tag : pip install git+https://github.com/huggingface/[email protected] and main: pip install git+https://github.com/huggingface/transformers worked successfully without errors. Could you please refer to this gist file.

Thanks @lysandre for resolving it