AxisMeru/prabhasa-nyaya-element-judge-4b-v0
AxisMeru/prabhasa-nyaya-element-judge-4b-v0 is a 4 billion parameter Qwen3-4B fine-tuned model designed for element-level legal judgment on Indian statute scenarios. It determines whether statutory elements are established or not_established based on provided facts, achieving 0.886 not_established recall on negative cases. This model specializes in structured legal element analysis for 14 specific Indian Penal Code and related statutes.
Loading preview...
Overview
AxisMeru/prabhasa-nyaya-element-judge-4b-v0 is a specialized 4 billion parameter model, fine-tuned from Qwen/Qwen3-4B, for performing element-level legal judgments on Indian statute scenarios. Its primary function is to analyze a given statute text, a scenario, and extracted facts to decide if each statutory element is established (supported by a specific fact) or not_established.
Key Capabilities
- Structured Legal Element Analysis: Provides a structured output indicating whether an element is established with a fact span or not.
- Specific Statute Coverage: Covers 14 Indian Penal Code and related statutes, including IPC §§ 302, 304A, 376, 378, 379, 380, 384, 390, 392, 405, 415, 420; NDPS Act § 22; and Prevention of Corruption Act § 7.
- High Performance on Negatives: Achieves a notable 0.886 not_established recall on negative cases, significantly outperforming Qwen3-4B-Base (0.125) and Qwen3-4B (0.224) baselines without few-shot prompting.
Training and Evaluation
The model was trained using supervised fine-tuning (SFTTrainer) on 1,463 element-judgment rows, with minority class oversampling. Evaluation on a 1,519-item held-out set showed an overall accuracy of 0.921. It is intended as a research prototype and not for actual legal decisions, with performance untested on statutes outside its covered 14.
Output Format
Outputs follow a specific format:
established <fact_id>:<start>:<end>
not_established