search

microsoft / LoRA

Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"
https://arxiv.org/abs/2106.09685
MIT License
9.76k stars 622 forks source link

Questions about running the cola dataset script #176

Open dengxingzhi opened 1 month ago

dengxingzhi commented 1 month ago

hello,An error occurs when I run the debert cola.sh script Traceback (most recent call last): File "examples/text-classification/run_glue.py", line 626, in main() File "examples/text-classification/run_glue.py", line 558, in main train_result = trainer.train(resume_from_checkpoint=checkpoint) File "/home/dengxingzhi/LoRA/examples/NLU/src/transformers/trainer.py", line 881, in train self._load_state_dict_in_model(state_dict) File "/home/dengxingzhi/LoRA/examples/NLU/src/transformers/trainer.py", line 1987, in _load_state_dict_in_model if set(load_result.missing_keys) == set(self.model._keys_to_ignore_on_save): TypeError: 'NoneType' object is not iterable

so,Is it a code problem,because running other scripts works fine

Zhoutong-Wu commented 1 month ago

I met the same problem, and I added the following in the _load_state_dict_in_model() function in transformers/trainer.py if self.model._keys_to_ignore_on_save == None: self.model._keys_to_ignore_on_save = [] Then I could run the cola.sh script.