Query!! Kindly let me know the alternative.

#48
by tusharpaul - opened

For Fine-Tuning (Using PEFT) code :

from peft import prepare_model_for_kbit_training

model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)

Error :

ValueError Traceback (most recent call last)
in <cell line: 3>()
1 from peft import prepare_model_for_kbit_training
2
----> 3 model.gradient_checkpointing_enable()
4 model = prepare_model_for_kbit_training(model)

/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py in gradient_checkpointing_enable(self)
1629 """
1630 if not self.supports_gradient_checkpointing:
-> 1631 raise ValueError(f"{self.class.name} does not support gradient checkpointing.")
1632 self.apply(partial(self._set_gradient_checkpointing, value=True))
1633

ValueError: PhiForCausalLM does not support gradient checkpointing.

Kindly let me know how to solve this issue.

Thank You

Email ID - tushar.paul@kusho.co

I also meet this issue. And i do not know how to solve it.

I got this -> https://huggingface.co/microsoft/phi-2/discussions/12
Maybe in future, it can help.

Microsoft org

Hello @ilovestudy and @tusharpaul .

We will proceed with the integration of Phi directly in transformers and it will fix the gradient checkpointing issue.

gugarosa changed discussion status to closed

Sign up or log in to comment