757 add exaone path 2.0

[cjs6211]

Fixes #757 .

Description

This PR adds the latest version of the EXAONE Path model, EXAONE Path 2.0.
The model is based on Hugging Face and is hosted at: https://huggingface.co/LGAI-EXAONE/EXAONE-Path-2.0.

Status

Ready

Please ensure all the checkboxes:

• Codeformat tests passed locally by running ./runtests.sh --codeformat.
• In-line docstrings updated.
• Update version and changelog in metadata.json if changing an existing bundle.
• Please ensure the naming rules in config files meet our requirements (please refer to: CONTRIBUTING.md).
• Ensure versions of packages such as monai, pytorch and numpy are correct in metadata.json.
• Descriptions should be consistent with the content, such as eval_metrics of the provided weights and TorchScript modules.
• Files larger than 25MB are excluded and replaced by providing download links in large_file.yml.
• Avoid using path that contains personal information within config files (such as use /home/your_name/ for "bundle_root").

[Copilot]

Pull Request Overview

This PR adds the latest version of the EXAONE Path model (EXAONE Path 2.0) to the Hugging Face models directory. The model is a pathology foundation model with 144 million parameters designed for whole-slide image (WSI) analysis and biomarker prediction.

• Adds complete EXAONE Path 2.0 model implementation with three-stage Vision Transformer architecture
• Includes comprehensive utilities for WSI processing, tissue segmentation, and tensor operations
• Provides sample data, documentation, and licensing for the new pathology model

Reviewed Changes

Copilot reviewed 11 out of 14 changed files in this pull request and generated 15 comments.

Show a summary per file

File	Description
hf_models/README.md	Updates model listing table to include EXAONE Path 2.0 entry
hf_models/EXAONE-Path-2.0/exaonepath.py	Main model implementation with three-stage inference pipeline
hf_models/EXAONE-Path-2.0/networks/vit.py	Vision Transformer architectures for the three processing stages
hf_models/EXAONE-Path-2.0/utils/wsi_utils.py	Utilities for whole-slide image processing and tissue segmentation
hf_models/EXAONE-Path-2.0/utils/tensor_utils.py	Tensor manipulation utilities for multi-stage processing
hf_models/EXAONE-Path-2.0/requirements.txt	Dependencies specification including CUDA-enabled PyTorch
hf_models/EXAONE-Path-2.0/config.json	Model configuration parameters
hf_models/EXAONE-Path-2.0/README.md	Comprehensive documentation and usage instructions
hf_models/EXAONE-Path-2.0/LICENSE	Custom license agreement for the model
hf_models/EXAONE-Path-2.0/pytorch_model.bin	Pre-trained model weights (LFS tracked)
hf_models/EXAONE-Path-2.0/samples/sample.svs	Sample whole-slide image file (LFS tracked)

Comments suppressed due to low confidence (2)

[ericspod]

Hi @cjs6211 thank you for the contribution. We do require that models in our zoo use the bundle format. You have a lot of your own code here so I wouldn't suggest reformatting everything to use MONAI components, but there does need to be at least an inference config file which can call into code you have defined. This bundle here is an example of an inference script with few components, and just calls into a defined method to implement an inference process. You can follow that pattern here but please do look at other bundles and see what you can do to make your code into a bundle.

We do also require that this model be in the models directory and not hf_models which contains information on those models hosted on Huggingface. We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern.

@aylward could you please double check that this model's license is acceptable to us, I think it's fine from our previous discussions on licenses.

Thanks!

[cjs6211]

Hi @cjs6211 thank you for the contribution. We do require that models in our zoo use the bundle format. You have a lot of your own code here so I wouldn't suggest reformatting everything to use MONAI components, but there does need to be at least an inference config file which can call into code you have defined. This bundle here is an example of an inference script with few components, and just calls into a defined method to implement an inference process. You can follow that pattern here but please do look at other bundles and see what you can do to make your code into a bundle.

We do also require that this model be in the models directory and not hf_models which contains information on those models hosted on Huggingface. We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern.

@aylward could you please double check that this model's license is acceptable to us, I think it's fine from our previous discussions on licenses.

Thanks!

Hi, @ericspod.
Regarding the comment "We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern", this actually aligns well with our original intention.

We have already uploaded the EXAONE Path 2.0 model to Hugging Face at https://huggingface.co/LGAI-EXAONE/EXAONE-Path-2.0 (different from the past model), and we would like this version to be referenced in hf_models following the same pattern as the existing EXAONE Path model.

However, since we were unsure of the correct process to request this kind of update, we submitted the current pull request. Please let us know exact way to refer to EXAONE Path 2.0.

Thank you again!

[ericspod]

Hi @cjs6211 thank you for the contribution. We do require that models in our zoo use the bundle format. You have a lot of your own code here so I wouldn't suggest reformatting everything to use MONAI components, but there does need to be at least an inference config file which can call into code you have defined. This bundle here is an example of an inference script with few components, and just calls into a defined method to implement an inference process. You can follow that pattern here but please do look at other bundles and see what you can do to make your code into a bundle.
We do also require that this model be in the models directory and not hf_models which contains information on those models hosted on Huggingface. We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern.
@aylward could you please double check that this model's license is acceptable to us, I think it's fine from our previous discussions on licenses.
Thanks!

Hi, @ericspod. Regarding the comment "We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern", this actually aligns well with our original intention.

We have already uploaded the EXAONE Path 2.0 model to Hugging Face at https://huggingface.co/LGAI-EXAONE/EXAONE-Path-2.0 (different from the past model), and we would like this version to be referenced in hf_models following the same pattern as the existing EXAONE Path model.

However, since we were unsure of the correct process to request this kind of update, we submitted the current pull request. Please let us know exact way to refer to EXAONE Path 2.0.

Thank you again!

Hi @cjs6211 I would suggest removing the files that you've proposed here and instead create a directory in hf_models with the contents similar to the existing model. You will want to replace the contents of the README.md and metadata.json files to match what your new model has but otherwise follow this as a template. The "model" link should point to the Huggingface link instead of the Github page however. I think that will work for what we want to host here and we can review again once that's done. Thanks!

[cjs6211]

Hi @cjs6211 thank you for the contribution. We do require that models in our zoo use the bundle format. You have a lot of your own code here so I wouldn't suggest reformatting everything to use MONAI components, but there does need to be at least an inference config file which can call into code you have defined. This bundle here is an example of an inference script with few components, and just calls into a defined method to implement an inference process. You can follow that pattern here but please do look at other bundles and see what you can do to make your code into a bundle.
We do also require that this model be in the models directory and not hf_models which contains information on those models hosted on Huggingface. We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern.
@aylward could you please double check that this model's license is acceptable to us, I think it's fine from our previous discussions on licenses.
Thanks!

Hi, @ericspod. Regarding the comment "We already reference a EXAONEPath model hosted on hf in hf_models, if this is a later version of the same model perhaps it should be hosted in hf instead of here and we add a reference in hf_models in a similar pattern", this actually aligns well with our original intention.
We have already uploaded the EXAONE Path 2.0 model to Hugging Face at https://huggingface.co/LGAI-EXAONE/EXAONE-Path-2.0 (different from the past model), and we would like this version to be referenced in hf_models following the same pattern as the existing EXAONE Path model.
However, since we were unsure of the correct process to request this kind of update, we submitted the current pull request. Please let us know exact way to refer to EXAONE Path 2.0.
Thank you again!

Hi @cjs6211 I would suggest removing the files that you've proposed here and instead create a directory in hf_models with the contents similar to the existing model. You will want to replace the contents of the README.md and metadata.json files to match what your new model has but otherwise follow this as a template. The "model" link should point to the Huggingface link instead of the Github page however. I think that will work for what we want to host here and we can review again once that's done. Thanks!

Hi, @ericspod. Thanks for the guidance. I’ve removed the directory under hf_models and created a new exaonepath_2.0 directory instead. I followed the structure of the existing model, added the necessary files, and updated the contents of README.md and metadata.json to reflect the new model. Everything has been uploaded and should now be ready for review.

Best,

[ericspod]

I'm good with this addition now, thanks!

[yiheng-wang-nv]

/build

[yiheng-wang-nv]

/build