Unable to prepare model for kbit training

🤗Transformers
1

Hi, i am trying to finetune phi 1.5 on a custom dataset using QLora. When i run:
model = prepare_model_for_kbit_training(model)
I receive the following error
AttributeError: ‘MixFormerSequentialForCausalLM’ object has no attribute ‘_set_gradient_checkpointing’
peft version : 0.5.0
transformer: 4.33.2

2 Likes
2

Same problem here, Because of that the GPU is filling up like crazy. If you got it solved by now, Could you share here?