Exploring Lightweight Large Language Models for Court View Generation

2026-05-19 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, medium

Summary

A study systematically explores the capabilities of lightweight Large Language Models (LLMs) (smaller than 2B parameters) for Criminal Court View Generation (CVG) and their impact on charge prediction. The research investigates how different LLM architectures and sizes affect CVG quality and charge prediction, compares lightweight LLMs with Deep Neural Networks (DNNs) in these tasks, and analyzes the effectiveness of predicting charges via court view generation versus direct prediction. The authors developed CVGEvalKit, an evaluation framework incorporating three public datasets (C3VG, LCVG, CCVG) for CVG and charge prediction tasks. Experiments involved training models on a mixed dataset and evaluating them on individual test sets, revealing insights into model architecture, size, and task interdependencies, underscoring the potential of lightweight LLMs in judicial AI.

Key takeaway

For Research Scientists developing legal AI systems, this study suggests that lightweight LLMs offer a promising avenue for automating Criminal Court View Generation and improving charge prediction accuracy. You should consider fine-tuning models like Qwen3-1.7B or InternLM2.5-1.8B-Chat, as they demonstrate strong performance, and explore generating court views prior to charge prediction to enhance overall system accuracy and interpretability.

Key insights

Lightweight LLMs show significant potential for Criminal Court View Generation and charge prediction in legal AI.

Principles

• LLM size correlates with improved scores post-fine-tuning.
• CVG quality directly impacts downstream charge prediction accuracy.

Method

Models are fine-tuned using Low-Rank Adaptation (LoRA) on a mixed CVG dataset, then evaluated on separate test sets for CVG and charge prediction.

In practice

• Use LoRA for efficient fine-tuning of LLMs in legal AI.
• Consider lightweight LLMs for complex legal reasoning tasks.

Topics

• Criminal Court View Generation
• Lightweight LLMs
• Charge Prediction
• Legal AI
• CVGEvalKit

Code references

• ZhitianHou/CVGEvalKit

Best for: Research Scientist, AI Scientist, NLP Engineer, Domain Expert

Related on AIssential

• VLegal-Bench: Cognitively Grounded Benchmark for Vietnamese Legal Reasoning of Large Language Models — VLegal-Bench offers a cognitively-grounded benchmark for LLMs in Vietnamese civil law, revealing specialized models outperform general ones on complex legal reasoning.
• Kirkland Hints It Could Fine-Tune LLMs For Own Legal AI Model — Large law firms are building custom, on-premise legal AI for differentiation and enhanced data privacy.
• Democratizing Legal Analytics: Resource-Efficient Information Extraction for Brazilian Case Law — Resource-efficient LLM fine-tuning enables scalable legal analytics in low-resource languages using legacy datasets.
• TW-LegalBench: Measuring Taiwanese Legal Understanding — LLMs demonstrate proficiency in legal qualification exams but struggle with precise statutory citation and advanced reasoning in civil-law contexts.
• Increase Recommendation Systems’ Precision with LLMs, Using Python — Optimize LLM-powered recommendation systems by combining cheap, high-recall filtering with expensive, high-precision reranking.
• An Empirical Study of Multi-Generation Sampling for Jailbreak Detection in Large Language Models — Multi-generation sampling is crucial for accurately assessing LLM jailbreak vulnerability, as single outputs underestimate risk.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.