Skip to content
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

text encoder(google/mt5-xxl) can be substituted by a smaller one? #550

Open
hongfengvvv opened this issue Dec 4, 2024 · 5 comments
Open

Comments

@hongfengvvv
Copy link

The file size of text encoder(google/mt5-xxl) model is more than 50GB. Is it possible to replace this model with a smaller one?
Or, will there be a smaller version of this model available in the future?

@sguo112
Copy link

sguo112 commented Dec 4, 2024

You can try mt5-small

@huipengzhang
Copy link

When I test Open-Sora-Plan1.3 use mt5-small ,it will make a error for this
error
I Identified the hidden_size of mt5-small is 512. It may causing shape mismatch between input and weight.How should we use mt5-small to inference,and do you need to modify any configurations? @sguo112

@sguo112
Copy link

sguo112 commented Dec 4, 2024

I haven't succeeded yet@huipengzhang

@yunyangge
Copy link
Collaborator

Hello, by default, smaller versions of mT5 cannot be used. Because their output dimensions are different from mT5-XXL, this results in a mismatch between DiT's weight dimensions and theirs.

@SFTJBD
Copy link

SFTJBD commented Dec 19, 2024

Hi, can DeepFloyd/t5-v1_1-xxl be used? I've found that's the default setting.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants