Description
Importing Gemma3ForMultimodalLM from gemma_pytorch.gemma.gemma3_model fails with:
ModuleNotFoundError: No module named 'gemma'
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
/tmp/ipykernel_2276/2279190376.py in <cell line: 0>()
1 from gemma_pytorch.gemma.config import get_model_config
----> 2 from gemma_pytorch.gemma.gemma3_model import Gemma3ForMultimodalLM
3
4 import os
5 import torch
/content/gemma_pytorch/gemma/gemma3_model.py in <module>
22 from typing import Any, List, Sequence, Tuple, Union
23
---> 24 from . import model as gemma_model
25 from . import config as gemma_config
26 from . import gemma3_preprocessor
/content/gemma_pytorch/gemma/model.py in <module>
22 from typing import Any, List, Optional, Sequence, Tuple, Union, Mapping
23
---> 24 from gemma import config as gemma_config
25 from gemma import tokenizer
26
ModuleNotFoundError: No module named 'gemma'
---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.
To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------
## Steps to reproduce
In a notebook/cloud kernel environment:
```python
from gemma_pytorch.gemma.config import get_model_config
from gemma_pytorch.gemma.gemma3_model import Gemma3ForMultimodalLM
Actual behavior
The import fails with:
ModuleNotFoundError: No module named 'gemma'
Full traceback:
from gemma_pytorch.gemma.config import get_model_config
from gemma_pytorch.gemma.gemma3_model import Gemma3ForMultimodalLM
# ...
# /content/gemma_pytorch/gemma/model.py in <module>
# ---> 24 from gemma import config as gemma_config
# 25 from gemma import tokenizer
#
# ModuleNotFoundError: No module named 'gemma'
Expected behavior
Gemma3ForMultimodalLM should import successfully when using the package namespace:
from gemma_pytorch.gemma.gemma3_model import Gemma3ForMultimodalLM
Suggested fix
In gemma/model.py, change the absolute imports:
from gemma import config as gemma_config
from gemma import tokenizer
to relative imports:
from . import config as gemma_config
from . import tokenizer
This matches how gemma3_model.py already imports sibling modules:
from . import model as gemma_model
from . import config as gemma_config
from . import gemma3_preprocessor
Environment
- Environment: Cloud notebook/kernel
- Package path used:
gemma_pytorch.gemma
- Python: 3.13
Description
Importing
Gemma3ForMultimodalLMfromgemma_pytorch.gemma.gemma3_modelfails with:Actual behavior
The import fails with:
Full traceback:
Expected behavior
Gemma3ForMultimodalLMshould import successfully when using the package namespace:Suggested fix
In
gemma/model.py, change the absolute imports:to relative imports:
This matches how
gemma3_model.pyalready imports sibling modules:Environment
gemma_pytorch.gemma