Pytorch Finetuning #6
Conversation
|
Hello. First, Thank you for your sincere pull request! |
|
I read your code quickly and I have some questions. |
train.py
Outdated
| def load_model(model, state_dict, device): | ||
| model = load_weight(model, state_dict) | ||
| model.to(device) | ||
| model.eval() |
There was a problem hiding this comment.
@jjbrophy47 Why did you try model.eval()?
Didn't you mean fine tuning?
There was a problem hiding this comment.
Ooops! That line is a mistake and should be deleted, thanks.
GPT2/data.py
Outdated
| else: | ||
|
|
||
| # Plain text | ||
| with open(path, 'r') as fp: |
There was a problem hiding this comment.
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 43: ordinal not in range(128)UnicodeDecodeError is occured in here. I think open with utf-8 is right.
There was a problem hiding this comment.
Ok I think that would be better, thanks.
|
branch name |
|
Thanks! I will also reflect the modifications to the master branch. |
Hi, I really like this repository and how easy it use to use as a PyTorch alternative for GPT2. In this pull request, I've added the ability to fine-tune a pre-trained gpt2 model in PyTorch. I've adapted training code from nshepperd: https://github.com/nshepperd/gpt-2/blob/finetuning/train.py
I hope you find this useful! Let me know if you have any questions or concerns!
-Jonathan Brophy