-
Notifications
You must be signed in to change notification settings - Fork 61
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
Question on Backbone for Experiments #27
Comments
Thank you for your interest and question. We did have some trainable LM baselines like BERT NCL/ONE/MTL. However, you're correct in noting that we didn't have more baselines specifically designed for trainable BERT. The primary reason for this is that many baselines (e.g., HAT) were not originally intended for LM, let alone trainable LM. The most straightforward approach to using them for LM would likely involve employing adapters or other parameter-efficient tuning methods (where the LM is fixed). We may add more baselines that are with trainable LM later. Before that, if you're interested in trainable LM baselines, feel free to check our latest paper and code in the repository https://github.com/UIC-Liu-Lab/ContinualLM. Our focus of that repository is on trainable LM, particularly in the pre-training setting where training the full LM is typically required. |
Ah I see. Thank you for your response. |
By the way, I have another question regarding this sentence "The most straightforward approach to using them for LM would likely involve employing adapters or other parameter-efficient tuning methods (where the LM is fixed)." Do you have any papers or other references about this? I am interested to know more about it. Thank you! |
Hi! I have a question after seeing your Performance section on README. Why doesn't it show the result for continual learning models using BERT (without frozen)? Thanks in advance!
The text was updated successfully, but these errors were encountered: