AttributeError: 'LlamaModel' object has no attribute 'model'

[sert121]

Facing the following error as soon as I start training the model using unsloth, tried a different environment, also reinstalling unsloth in a fresh env, nothing works, could you please help out?

0%| | 0/8820 [00:00<?, ?it/s]Traceback (most recent call last):
line 98, in train
trainer_stats = trainer.train()
File "", line 157, in train
File "", line 331, in _fast_inner_training_loop
File "/network/scratch/y/yash.more/rehaul_textmin/v1/lib/python3.10/site-packages/unsloth/models/_utils.py", line 1010, in _unsloth_get_batch_samples
f = model.base_model.model.forward if hasattr(model, "base_model") else model.forward
File "/network/scratch/y/yash.more/rehaul_textmin/v1/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1931, in getattr
raise AttributeError(
AttributeError: 'LlamaModel' object has no attribute 'model'

[danielhanchen]

Did you run model = FastLanguageModel.get_peft_model

[sert121]

I didnt, but i dont wanna use peft,I wanted to do full finetuning which was working fine before with unsloth

[sert121]

Could you give suggestions to fix/debug it further? i only removed the peft part, but the training script is the same as suggested by unsloth

[danielhanchen]

@sert121 Apologies currently does not yet support full finetuning. We plan to add support soon! Sorry!

[sert121]

@danielhanchen I actually had finetuned (fully) using unsloth few weeks ago! I just had to remove the prft config and use the fastmodel in sft. But that somehow stopped working recently, could you shed some light on it?

[danielhanchen]

@sert121 Oh it's not supposed to work :( I added better error handling - if you want, you could try downgrading Unsloth to an earlier version maybe?

[sert121]

How do i do that? it downloads automatically the newer oackages whenever it loads unsloth, any suggestions?