emooreog/DAEMON-bert-debiased

DAEMON BERT Debiased Model

This model is part of the DAEMON (Debiasing and Analysis for Enhanced Model Operations Network) project, which aims to reduce gender and racial bias in language models.

Model Description

• Model Type: BERT (fine-tuned)
• Task: Debiased text classification
• Training Data: Custom dataset with balanced gender and racial examples
• Purpose: Reduce harmful biases while maintaining model performance

Usage

# Load model and tokenizer
model = AutoModel.from_pretrained("emooreog/DAEMON-bert-debiased")
tokenizer = AutoTokenizer.from_pretrained("emooreog/DAEMON-bert-debiased")

# Example usage
inputs = tokenizer("Example text to analyze", return_tensors="pt")
outputs = model(**inputs)

Training Details

This model was fine-tuned using a curriculum learning approach with carefully crafted examples to help reduce gender and racial biases.

Limitations

While this model has been trained to reduce certain biases, it may still exhibit biased behavior in some contexts. Always evaluate the model's outputs critically.

Citation

If you use this model in your research, please cite:

@misc{DAEMON2025,
  author = {Moore, Eddie},
  title = {DAEMON: Debiasing and Analysis for Enhanced Model Operations Network},
  year = {2025},
  publisher = {GitHub},
  url = {https://github.com/mooreem91/DAEMON}
}

Model Information

{
  "training_completed": "2025-03-13T15:15:36.285312",
  "model_purpose": "Debiased BERT for DA\u039eMON framework",
  "original_data_metadata": {
    "generated": "2025-03-13T15:10:35.550534",
    "num_examples": 365,
    "purpose": "BERT model debiasing",
    "categories": [
      "gender",
      "racial"
    ],
    "balance_info": {
      "template_weights": {
        "professional": 1.5,
        "academic": 1.5,
        "achievement": 1.3,
        "leadership": 1.0,
        "occupation": 1.0
      },
      "metrics_included": true,
      "total_templates": {
        "gender": 3,
        "racial": 3
      }
    }
  },
  "debiasing_focus": [
    "gender",
    "racial"
  ],
  "model_version": "1.0.0-debiased"
}