-
Notifications
You must be signed in to change notification settings - Fork 1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error when Using 8-bit Quantization #1616
Comments
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. |
this still needs to be addressed |
Hi |
Hi @younesbelkada it is |
thanks! Can you print the model after merging it ? Alternatively can you share a model that we can look into on the Hub? |
I am encountering a data type mismatch error when using 8-bit quantization with the PEFT library and SFTTrainer for fine-tuning a language model. The error occurs during the generation phase after loading the fine-tuned model.
Here's an overview of my workflow:
Here's the code snippet for loading the fine-tuned model:
The error message I'm getting is:
I have also printed the data types of the model parameters, and they appear to be a mix of torch.float16 and torch.int8, which is expected when using 8-bit quantization.
I would appreciate any guidance why am facing this issue
NOTE: this is you did not appear to me when I load the model in 4bits but after the fine-tuning I want to use the model in 8bits to get better accuracy (please correct me if my hypothesis is not correct)
The text was updated successfully, but these errors were encountered: