snowsadh/themis-judge-3b
Themis Judge 3B by snowsadh is a 3 billion parameter Llama 3.2 Instruct fine-tune, specifically optimized to act as a judge in moot court simulations. It evaluates legal arguments across multiple dimensions, generates structured judicial responses, manages speaker transitions, and maintains internal notes. This model excels at providing consistent, structured feedback for legal argumentation within a courtroom context.
Loading preview...
Themis Judge 3B: Specialized Moot Court AI
Themis Judge 3B, developed by snowsadh, is a fine-tuned Llama 3.2 3B Instruct model designed for moot court simulation environments. Its primary function is to act as a presiding judge, evaluating legal arguments and managing courtroom proceedings.
Key Capabilities
- Structured Argument Evaluation: Assesses legal arguments based on criteria like Legal Application, Issue Relevance, Argument Flow, and Bench Handling, providing scores from -3 to +3.
- Judicial Response Generation: Produces courtroom-style responses and questions.
- Speaker Management: Decides speaker transitions based on argument flow and judicial intervention.
- Internal Judge Notes: Generates private notes for each turn, reflecting the judge's internal assessment.
- High Structured Output Reliability: Achieved 0 parse failures out of 51 evaluations, compared to 51/51 for the base Llama 3.2 3B model, demonstrating robust schema adherence.
Training and Performance
The model was fine-tuned using QLoRA on a custom dataset of 504 examples from 56 constitutional law cases, focusing on multi-turn legal argumentation. Evaluation showed an average Mean Absolute Error (MAE) of 1.0589 across scoring dimensions, indicating moderate alignment with gold standard scores. Training and evaluation losses remained closely aligned, suggesting stable convergence without significant overfitting.
Good For
- Interactive Moot Court Simulators: Ideal for applications requiring an AI to simulate a judge's role.
- Legal Argumentation Training: Provides structured feedback for students practicing legal arguments.
- Research in Legal AI: A specialized model for exploring AI's role in legal reasoning and evaluation.
Limitations
- Primarily focused on Indian constitutional law contexts.
- May generate legally incorrect reasoning and is not suitable for legal advice or real-world judicial decision-making.