TauRecoID

Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

This modelcard aims to be a base template for new models. It has been generated using this raw template.

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

• Developed by: Joschka Birk, Anna Hallin, Gregor Kasieczka
• Model type: Transformer
• Language(s) (NLP): Pytorch
• Finetuned from model: https://doi.org/10.1088/2632-2153/ad66ad

The OmniJet- \(\alpha\) model was published in here was used as the base model for identifying hadronically decaying taus, reconstructing their kinematics and predicting their decay mode. The base model, initially trained on JetClass dataset, was now fine-tuned on Fu \(\tau\)ure dataset. The models included here are for 3 separate tasks:

• Tau-tagging (binary classification)
• Tau kinematic reconstruction (regression)
• Tau decay mode classification (multiclass-classification)

And for 3 different ways of training:

• From scratch
• Fixed backbone (fine-tune only head)
• Fine-tuning (fine-tune both head and backbone)

This will add up to 9 different models.

Model Sources [optional]

<!-- Provide the basic links for the model. -->

• Repository (base model): https://github.com/uhh-pd-ml/omnijet_alpha
• Repository (fine-tuned model): https://github.com/HEP-KBFI/ml-tau-en-reg
• Paper: https://doi.org/10.1088/2632-2153/ad66ad

Uses

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

The intended use of the models is to study the feasibility of foundation models for the purposes of reconstructing and identifying hadronically decaying tau leptons.

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

This model is not intended for physics measurements on real data. The trainings have been done on CLIC detector simulations.

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

The model has only been trained on simulation data and has not been validated against real data. Although the base model has been published in a peer-reviewed journal, the fine-tuned model has not been.

How to Get Started with the Model

Use the code below to get started with the model.

# Clone the repository
git clone [email protected]:HEP-KBFI/ml-tau-en-reg.git --recursive
cd ml-tau-en-reg


# Get the models
git clone https://huggingface.co/LauritsT/TauRecoID models

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

The data used to fine-tune the base model can be found here: Fu \(\tau\)ure dataset

Training Hyperparameters

• No hyperparameter tuning has been done. <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->

Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

Training on 1M jets on AMD MI250x for 100 epochs takes ~8h.

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data, Factors & Metrics

Testing Data

<!-- This should link to a Dataset Card if possible. -->

Testing data can also be found in the same Zenodo entry as the rest of the data.

Software

Software to train and analyze the model

Citation [optional]

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

OmniJet- \(\alpha\)

Model Card Authors [optional]

Laurits Tani ([email protected])

Model Card Contact

Laurits Tani ([email protected])