Update model contribution guide #2254

divyashreepathihalli · 2025-05-16T20:27:17Z

Since KerasHub has evolved, updating the model contribution guides. This will make contributions consistent and lower review time.
Markdown preview here
https://github.com/divyashreepathihalli/keras-nlp/blob/contributing_guide/CONTRIBUTING_MODELS.md

mattdangerw

Thanks! Left some comments.

mattdangerw · 2025-05-19T19:33:29Z

CONTRIBUTING_MODELS.md


 - [ ] Open an issue or find an issue to contribute a backbone model.

-### Step 2: PR #1 - Add XXBackbone
+### Step 2: PR #1 - Model Folder


Do we want this many PRs? Might be better to ask for a single PR with backbone, initial task, and colab showing usage and results. Less likely to have incomplete model contributions.

I'd say definitely not this, we don't want people opening up PRs just to create empty model folders. That is just more review for us (with nothing of value in the PR).

mattdangerw · 2025-05-19T19:34:17Z

CONTRIBUTING_MODELS.md


-### Step 4: PR #3 - Add XX Presets
+### Step 5: PR #3 - Add `XX` Tasks and Preprocessors (Optional)


We might want to consider saying at least one task is not optional.

mattdangerw · 2025-05-19T19:35:14Z

CONTRIBUTING_MODELS.md


-#### Unit Tests
+[Example](https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/distil_bert/distil_bert_backbone.py#L187-L189)


You are adding a lot of outdated code links that no longer work. Please check all these!

mattdangerw · 2025-05-19T19:36:48Z

CONTRIBUTING_MODELS.md

-and return the dictionary in the form expected by the model.
+- New Task Models (e.g., TokenClassifier, ImageSegmentation)
+- Parameter-Efficient Fine-Tuning (LoRA support)
+- Quantization (QLoRA support)


what do we expect to be added for lora and qlora support? this seems kinda ill defined

mattdangerw · 2025-05-19T19:39:27Z

CONTRIBUTING_MODELS.md

+- [ ] Add `xx/xx_presets.py` with links to weights uploaded to Kaggle KerasHub  
+      [Example](https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/distil_bert/distil_bert_presets.py)
+
+- [ ] Stage the model presets on KerasHub’s [Kaggle org page](https://www.kaggle.com/organizations/kerashub) using this [invite link](https://kaggle.com/organizations/kerashub/invite/c4b8baa532b8436e8df8f1ed641b9cb5)


We might not want to make this invite link public. Won't anyone be able to join the org with this? What kind of permissions does this get you (model creation? model deletion?).

mattdangerw · 2025-05-19T19:41:47Z

CONTRIBUTING_MODELS.md


-### Step 4: PR #3 - Add XX Presets
+#### Checkpoint Conversion Script (tools/checkpoint_conversion/convert_your_model_checkpoints.py)


We should mention timm/huggingface converters. And show one of those as the primary example.

Basically, we should say that is our preferred mode of checkpoint is to go from a supported conversion source (timm, transformers) and write a built in library converted. Then the convert checkpoint tool is just a thin wrapper around this converter.

Alternately (and more advanced) would be to write a converted from another format, directly in the tools/ script.

mattdangerw · 2025-05-19T19:44:07Z

CONTRIBUTING_MODELS.md

+
+##### Implementation
+
+- **Text**: `XXTokenizer`, subclassing from KerasHub tokenizers


Probably should link these code files (tokenizer, imageconverter, audioconverter).

divyashreepathihalli · 2025-07-11T00:07:16Z

/gemini review

gemini-code-assist

Code Review

This pull request updates the model contribution guide, providing a clearer and more comprehensive structure for contributors. Suggestions include fixing a potentially inaccessible link, correcting a typo, and improving section structure for enhanced clarity.

gemini-code-assist · 2025-07-11T00:08:29Z

CONTRIBUTING_MODELS.md

+
+- [ ] A Colab notebook link in the PR description that matches the outputs of
+the implemented backbone model with the original source
+      [Example](https://colab.sandbox.google.com/drive/1R99yFJCbxTEpcxFHa2RtlwQWahUIPCJC?usp=sharing)


The example Colab link points to colab.sandbox.google.com, which is typically used for internal Google sandboxes and may not be accessible to external contributors. To ensure everyone can view the example, please update this to a publicly accessible colab.research.google.com link.

gemini-code-assist · 2025-07-11T00:08:29Z

CONTRIBUTING_MODELS.md

+- Major changes: [`DebertaV3`](https://github.com/keras-team/keras-hub/tree/master/keras_hub/models/deberta_v3)
+- Minor tweaks: [Whisper attention layer](https://github.com/keras-team/keras-hub/pull/801/files#diff-8533ae3a7755c0dbe95ccbb71f85c677297f687bf3884fadefc64f1d0fdce51aR22)
+
+Do **not** include `from_presets()` in this PR.


There's a typo in the method name. The method is from_preset() (singular), but here it's written as from_presets() (plural). This is inconsistent with the KerasHub API and another mention in this guide (line 259). Correcting this will avoid confusion for contributors.

Suggested change

Do **not** include `from_presets()` in this PR.

Do **not** include `from_preset()` in this PR.

gemini-code-assist · 2025-07-11T00:08:30Z

CONTRIBUTING_MODELS.md

+- Provide a script that converts weights from their original format (e.g.,
+PyTorch .bin, TensorFlow SavedModel) to the Keras H5 format expected by KerasHub.
+- This script should be reusable and clearly documented.
+- It's crucial for verifying weight conversion accuracy and for future updates.
+End-to-End Validation Colab
+- This is the most important validation step.


The description for the Checkpoint Conversion Script is missing a period at the end of the sentence.

gemini-code-assist · 2025-07-11T00:08:30Z

CONTRIBUTING_MODELS.md

+End-to-End Validation Colab
+- This is the most important validation step.

-After wrapping up the preset configuration file, you need to
-add the `from_preset` function to all three classes, i.e., `DistilBertBackbone`,
-and `DistilBertTokenizer`. Here is an
-[example](https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/distil_bert/distil_bert_backbone.py#L187-L189).
+#### End-to-End Colab

-The testing for presets is divided into two: "large" and "extra large".
-For "large" tests, we pick the smallest preset (in terms of number of parameters)
-and verify whether the output is correct. For "extra large tests", we loop over
-all the presets and just check whether the backbone and the tokenizer can
-be called without any error.
+- Load task model using `from_preset()`
+- Run task (e.g., classification, generation)
+- Compare output with original model


This section has a redundant and confusing structure with two similar headings (End-to-End Validation Colab and End-to-End Colab). I suggest merging them into a single, clear section for the end-to-end validation Colab to improve readability.

Suggested change

End-to-End Validation Colab

- This is the most important validation step.

After wrapping up the preset configuration file, you need to

add the `from_preset` function to all three classes, i.e., `DistilBertBackbone`,

and `DistilBertTokenizer`. Here is an

[example](https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/distil_bert/distil_bert_backbone.py#L187-L189).

#### End-to-End Colab

The testing for presets is divided into two: "large" and "extra large".

For "large" tests, we pick the smallest preset (in terms of number of parameters)

and verify whether the output is correct. For "extra large tests", we loop over

all the presets and just check whether the backbone and the tokenizer can

be called without any error.

- Load task model using `from_preset()`

- Run task (e.g., classification, generation)

- Compare output with original model

#### End-to-End Validation Colab

This is the most important validation step.

- Load task model using `from_preset()`

- Run task (e.g., classification, generation)

- Compare output with original model

Update contributing guide

9b2692a

divyashreepathihalli marked this pull request as draft May 16, 2025 20:27

divyashreepathihalli changed the title ~~Update contributing guide~~ [WIP] Update contributing guide May 16, 2025

divyashreepathihalli added 2 commits May 16, 2025 22:43

update iteration keras-team#2

daaaf43

reformat

60928dc

divyashreepathihalli changed the title ~~[WIP] Update contributing guide~~ Update contributing guide May 16, 2025

divyashreepathihalli requested review from mattdangerw and abheesht17 May 16, 2025 23:01

divyashreepathihalli marked this pull request as ready for review May 16, 2025 23:02

divyashreepathihalli changed the title ~~Update contributing guide~~ Update model contribution guide May 16, 2025

nit

c24af31

mattdangerw reviewed May 19, 2025

View reviewed changes

remove split PR and invite link

4687b06

gemini-code-assist bot reviewed Jul 11, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Update model contribution guide #2254

Update model contribution guide #2254

Uh oh!

divyashreepathihalli commented May 16, 2025 •

edited

Loading

Uh oh!

mattdangerw left a comment

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

mattdangerw May 19, 2025

Uh oh!

divyashreepathihalli commented Jul 11, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

Uh oh!


		### Step 4: PR #3 - Add XX Presets
		### Step 5: PR #3 - Add `XX` Tasks and Preprocessors (Optional)


		#### Unit Tests
		[Example](https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/distil_bert/distil_bert_backbone.py#L187-L189)


		### Step 4: PR #3 - Add XX Presets
		#### Checkpoint Conversion Script (tools/checkpoint_conversion/convert_your_model_checkpoints.py)


		##### Implementation

		- Text: `XXTokenizer`, subclassing from KerasHub tokenizers

	Do not include `from_presets()` in this PR.
	Do not include `from_preset()` in this PR.

Update model contribution guide #2254

Are you sure you want to change the base?

Update model contribution guide #2254

Uh oh!

Conversation

divyashreepathihalli commented May 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mattdangerw left a comment

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

mattdangerw May 19, 2025

Choose a reason for hiding this comment

Uh oh!

divyashreepathihalli commented Jul 11, 2025

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

divyashreepathihalli commented May 16, 2025 •

edited

Loading