google

/

gemma-3-4b-it

like 174

Follow

Google 9.4k

PEFT integration needs vocab_size to operate.
To reproduce the error train a sample batch with peft.

[rank0]:     self.save_model(output_dir, _internal_call=True)
[rank0]:   File "/home/nmysore/workspace/reftml/venv/lib64/python3.9/site-packages/transformers/trainer.py", line 3873, in save_model
[rank0]:     self._save(output_dir)
[rank0]:   File "/home/nmysore/workspace/reftml/venv/lib64/python3.9/site-packages/transformers/trainer.py", line 3977, in _save
[rank0]:     self.model.save_pretrained(
[rank0]:   File "/home/nmysore/workspace/reftml/venv/lib64/python3.9/site-packages/peft/peft_model.py", line 326, in save_pretrained
[rank0]:     output_state_dict = get_peft_model_state_dict(
[rank0]:   File "/home/nmysore/workspace/reftml/venv/lib64/python3.9/site-packages/peft/utils/save_and_load.py", line 258, in get_peft_model_state_dict
[rank0]:     and (vocab_size != model.config.__class__.from_pretrained(model_id).vocab_size)
[rank0]:   File "/home/nmysore/workspace/reftml/venv/lib64/python3.9/site-packages/transformers/configuration_utils.py", line 214, in __getattribute__
[rank0]:     return super().__getattribute__(key)
[rank0]: AttributeError: 'Gemma3Config' object has no attribute 'vocab_size'

Proposed change:
Add vocab_size to config.json