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

Update PaliGemma2 presets #2004

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Update PaliGemma2 presets #2004

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion keras_hub/src/models/pali_gemma/pali_gemma_backbone.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,7 @@ class PaliGemmaBackbone(Backbone):
}

# Pretrained PaliGemma decoder.
model = keras_hub.models.PaliGemmaBackbone.from_preset("pali_gemma_mix_224")
model = keras_hub.models.PaliGemmaBackbone.from_preset("pali_gemma_3b_mix_224")
model(input_data)

# Randomly initialized PaliGemma decoder with custom config.
Expand Down
42 changes: 38 additions & 4 deletions keras_hub/src/models/pali_gemma/pali_gemma_presets.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
"""PaliGemma model preset configurations."""

import re

# Metadata for loading pretrained model weights.
backbone_presets = {
"pali_gemma_3b_mix_224": {
Expand Down Expand Up @@ -53,7 +55,7 @@
"kaggle_handle": "kaggle://keras/paligemma/keras/pali_gemma_3b_896/3",
},
# PaliGemma2
"pali_gemma2_3b_ft_docci_448": {
"pali_gemma2_ft_docci_3b_448": {
"metadata": {
"description": (
"3 billion parameter, image size 448, 27-layer for "
Expand All @@ -66,9 +68,10 @@
"path": "pali_gemma2",
"model_card": "https://www.kaggle.com/models/google/paligemma-2",
},
"kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_3b_ft_docci_448/1",
# TODO: Rename `pali_gemma_2_ft_docci_3b_448` to `pali_gemma2_ft_docci_3b_448`
"kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma_2_ft_docci_3b_448/1",
Copy link
Member

Choose a reason for hiding this comment

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

Does kaggle handle redirects with a rename? @divyashreepathihalli

Copy link
Member

Choose a reason for hiding this comment

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

This is probably important to resolve before we merge this, otherwise we need to start thinking about the ordering of things.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Unable to just rename, we would need to reupload.

},
"pali_gemma2_10b_ft_docci_448": {
"pali_gemma2_ft_docci_10b_448": {
"metadata": {
"description": (
"10 billion parameter, 27-layer for SigLIP-So400m vision "
Expand All @@ -81,7 +84,7 @@
"path": "pali_gemma2",
"model_card": "https://www.kaggle.com/models/google/paligemma-2",
},
"kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_10b_ft_docci_448/1",
"kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_ft_docci_10b_448/1",
},
"pali_gemma2_pt_3b_224": {
"metadata": {
Expand Down Expand Up @@ -219,3 +222,34 @@
"kaggle_handle": "kaggle://keras/paligemma2/keras/pali_gemma2_pt_28b_896/1",
},
}

# Ensure compatibility with the official naming convention.
Copy link
Member

Choose a reason for hiding this comment

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

I think we should keep this a lot simpler, just a list of aliases and resolved names. No regexes.

We should also figure out how we don't list the old name in class.presets, that will prevent them from being rendered on keras.io. We don't want to document the old names presumably.

Copy link
Member

Choose a reason for hiding this comment

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

Maybe we could add something to pali_gemma/__init__.py.

from keras_hub.src.utils.preset_utils import register_presets
from keras_hub.src.utils.preset_utils import register_aliases

# Old names for presets we no longer use but still support.
register_aliases([
    ("pali_gemma2_10b_ft_docci_448", "pali_gemma2_ft_docci_10b_448"),
])

# pali_gemma2_[3|10|28b]_[variant]_[image_size]
compatible_preset_names = []
for preset_name in backbone_presets.keys():
if re.match(r"pali_gemma2_(.+)_(.+)_(.+)_(.+)", preset_name):
# Ex: pali_gemma2_ft_docci_3b_448 -> pali_gemma2_3b_ft_docci_448
compatible_preset_names.append(
(
re.sub(
r"pali_gemma2_(.+)_(.+)_(.+)_(.+)",
r"pali_gemma2_\3_\1_\2_\4",
preset_name,
),
preset_name,
)
)
elif re.match(r"pali_gemma2_(.+)_(.+)_(.+)", preset_name):
# Ex: pali_gemma2_pt_3b_224 -> pali_gemma2_3b_pt_224
compatible_preset_names.append(
(
re.sub(
r"pali_gemma2_(.+)_(.+)_(.+)",
r"pali_gemma2_\2_\1_\3",
preset_name,
),
preset_name,
)
)
for compatible_preset_name, preset_name in compatible_preset_names:
backbone_presets[compatible_preset_name] = backbone_presets[preset_name]
Loading