Use fine tuned model

[seyyedjavadrazavi]

I fine-tuned the bert model on my pc, then used the bert-score -r comments.txt -c all_tags.txt --model my_model --num_layers 9 code to use the fine-tuned model on bert score, but this error happened.

[Tiiiger]

hi @seyyedjavadrazavi ,

This is saying you didn't give it the correct path to the model folder. Please check you can load your model correctly by

from transformers import AutoModel

AutoModel.from_pretrained("<your model>")

[seyyedjavadrazavi]

Thanks for the quick reply.
I am new to this field and to fine-tune the bert used #61, #12, and #82 and finally because I want to fine-tune the bert in an unsupervised way, I used this link for fine-tuning and torch.save(model, 'C:/Users/Seyyed Javad Razavi/Desktop/New folder (2)/f_tune_analyticsvidhya/model_2') code to save the fine-tuned model which saved just a file with the name 'model_2' without any config file.
Now the problem is the code you mentioned is not work for me and this error happened.

[Tiiiger]

Hi @seyyedjavadrazavi ,

you should use the huggingface save_pretrained method for this

[seyyedjavadrazavi]

Thanks, @Tiiiger.
I did that and load the fine-tuned model by model = AutoModel.from_pretrained("./new_model/model/", from_tf=True) and used the score function P, R, F1 = score(crnt_tg, tags_name, model_type = model, num_layers = 10, device="cuda:0"), but the error was :

[felixgwu]

Hi @seyyedjavadrazavi ,

The argument model_type should be a string, please try this:

P, R, F1 = score(crnt_tg, tags_name, model_type = "./new_model/model/", num_layers = 10, device="cuda:0")

[seyyedjavadrazavi]

@felixgwu
Thanks a lot.
It works now.