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

[mieb] Investigate voc2007 vlm2vec #1825

Draft
wants to merge 6 commits into
base: mieb
Choose a base branch
from

Conversation

isaac-chung
Copy link
Collaborator

@isaac-chung isaac-chung commented Jan 16, 2025

Fixes #1792

  • lrap scores increase with increasing samples_per_label
  • was able to get 72% for both lora and full (full slightly higher than lora) when samples_per_label=64.
  • using samples_per_label=64 yields 0.787 lrap for voyage-multimodal-3

Checklist

  • Run tests locally to make sure nothing is broken using make test.
  • Run the formatter to format the code using make lint.

Copy link
Collaborator Author

@isaac-chung isaac-chung left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@gowitheflow-1998 I've left some notes below to supplement findings. It seems like when LoRA was run, samples_per_label might have been low, e.g. 8, instead of 64.

Comment on lines +158 to +162
_unique_train_embeddings = normalize_embeddings_to_numpy(
model.get_image_embeddings(
unique_train_images,
**encode_kwargs,
)
Copy link
Collaborator Author

@isaac-chung isaac-chung Jan 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No change from adding normalize_embeddings_to_numpy. Will revert before merging.

Comment on lines +179 to +181
X_test = normalize_embeddings_to_numpy(
model.get_image_embeddings(test_images, **encode_kwargs)
)
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same here, will revert this change.

"languages": [
"eng-Latn"
],
"lrap": 0.7205710375157255,
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Using commits to track lrap scores for both models at different samples_per_label. Will remove before merging.

@@ -55,3 +55,5 @@ class VOC2007Classification(AbsTaskImageMultilabelClassification):

# To be removed when we want full results
n_experiments: int = 5

samples_per_label: int = 64
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Main change.

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

Successfully merging this pull request may close these issues.

1 participant