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.

Sign up for GitHub

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

Jump to bottom

port t5 and clip to nnx.Module #21

Open
SauravMaheshkar opened this issue Oct 15, 2024 · 0 comments
Open

port t5 and clip to nnx.Module #21

SauravMaheshkar opened this issue Oct 15, 2024 · 0 comments
Labels
feature 🚀 New feature or request

Comments

@SauravMaheshkar
Copy link
Collaborator

Currently, we load the t5 and clip models as torch modules and wrap their outputs as JAX arrays. This prevents OOM during inference (since we can use the .to() method); having a pure JAX inference pipeline would be ideal.

References

  1. Initialize Flax model params on CPU  huggingface/transformers#24711
  2. feat(flax): leave restored weights on CPU huggingface/transformers#15295
@SauravMaheshkar SauravMaheshkar added the feature 🚀 New feature or request label Oct 15, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
feature 🚀 New feature or request
Projects
JFLUX Roadmap
Status: Todo
Milestone
No milestone
Development

No branches or pull requests
1 participant
@SauravMaheshkar