---
language: en
license: apache-2.0
tags:
- bias-mitigation
- social-bias
- stereotypes
- fine-tuned
- llama
- causal-lm
datasets:
- stereoset
metrics:
- stereoset-score
- bias-metrics
base_model: HuggingFaceTB/SmolLM-360M
model-index:
- name: SmolLM-360M-Bias-Mitigated
  results:
  - task:
      type: bias-evaluation
      name: Social Bias Evaluation
    dataset:
      name: StereoSet
      type: stereoset
    metrics:
    - type: ss_score
      value: 0.5345
      name: StereoSet Score
    - type: bias_severity
      value: 0.0345
      name: Bias Severity
---

# SmolLM-360M Fine-tuned for Bias Mitigation

**Note:** _This project is intended for educational purposes only_

This model is a fine-tuned version of [HuggingFaceTB/SmolLM-360M](https://huggingface.co/HuggingFaceTB/SmolLM-360M) specifically trained to mitigate social biases including gender, profession, race, and religion stereotypes.

Developed for the Natural Language Processing Course (Academic Year 2024-25) as part of the Master Degree in Computer science at the Univeristy of Milan.

The main repository source code can be found here: https://github.com/giovannifil-64/nlp