StanfordSCALE

This organization hosts models developed through research projects led by contributors at the Stanford SCALE Initiative.

1. Overview

This organization contains models trained across a range of research efforts within SCALE. These models support diverse tasks—primarily classification and analysis—applied to domain-specific datasets.

Because many models are trained on specialized or proprietary datasets, their performance and applicability may not generalize beyond their original research context.

Repositories are organized into collections, each corresponding to a specific research project or intervention (e.g., Tutor Copilot).

2. Model Naming Convention

We use a consistent naming scheme with the following convention:

<project>-<task>-<base_model>-<size>-v<version>

Field Definitions

• project → research project or initiative (e.g., oym, tutor-copilot)
• task → task type (e.g., classification, qa, summarization)
• base_model → pretrained backbone (e.g., roberta, llama3, bert)
• size → model size (e.g., base, large, 7b)
• version → iteration of the same model setup (e.g., v1, v2), updated when data, training, or performance changes.

Examples:

For the OYM project, for a relationship management classifier with Roberta base model the name would be the following:

• oym-relationship-classification-roberta-base-v1

3. Documentation Standards

All models include:

• Training data details
• Hyperparameters
• Evaluation benchmarks
• Known limitations

All models must follow the SCALE model card template to ensure consistent, transparent, and reproducible documentation.

📘 Model Card Template:
Model Card Template

Model Card Template Instructions

4. Highlighted projects

{{}}

{{1 sentence description}} → 🤗 Model: {{link}}

{{Project Name}}

{{1 sentence description}} → 🤗 Model: {{link}}

5. Privacy

When creating a model repository set the model privacy setting to Public. This setting will allow other user to see this model. Only members of StanfordSCALE can commit.

6. Citation

Use the following citation convention

@misc{{ORG_NAME}_{MODEL}_{YEAR},
  title = {{TITLE}},
  author = {{AUTHORS}},
  year = {{YEAR}},
  url = {{URL}}
}

7. Project List

A list of the pretrained models, along with the tasks they are designed for, the datasets used, and the researcher in charge, can be found here.

8. Contact

• 📧 Contact information
  • JP Martinez: jpmc@stanford.edu
  • Hsiaolin Hsie: hsiaolin@stanford.edu
  • Ana Trindade Ribeiro: anactr@stanford.edu
• 🌐 https://scale.stanford.edu
• 💻 https://github.com/scale-nssa