Lshan-1.0 Technical Report

2025-03-10

AI Summary

Current legal large language models suffer from weak legal reasoning capabilities and insufficient domain-specific training data, limiting their effectiveness in high-precision judicial tasks such as procuratorial practice. To address these bottlenecks, this project introduces Lshan-1.0—the first-generation Chinese legal-domain reasoning LLM—designed through a novel expert-in-the-loop training paradigm involving legal professionals throughout model development. We construct a million-scale, province-wide (covering all 31 provinces) Chinese legal corpus spanning over 20 criminal offenses. Leveraging the DeepSeek-R1-Distilled architecture, Lshan-1.0 integrates supervised fine-tuning (SFT) with unsupervised large-scale reinforcement learning (RL) to enhance both legal reasoning fidelity and output interpretability. The model is released in three dense parameter configurations: 14B, 32B, and 70B. Empirical evaluation on real-world procuratorial tasks demonstrates substantial improvements in reasoning accuracy, logical rigor, and explainability over both general-purpose and existing legal LLMs—validated by expert human assessment.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

In this report, we introduce our first-generation reasoning model, Lshan-1.0, a large language model designed for the highly specialized Chinese legal domain, offering comprehensive capabilities to meet diverse realistic needs. Existing legal LLMs face two primary challenges. Firstly, their design and evaluation are predominantly driven by computer science perspectives, leading to insufficient incorporation of legal expertise and logic, which is crucial for high-precision legal applications, such as handling complex prosecutorial tasks. Secondly, these models often underperform due to a lack of comprehensive training data from the legal domain, limiting their ability to effectively address real-world legal scenarios. To address this, we first compile millions of legal documents covering over 20 types of crimes from 31 provinces in China for model training. From the extensive dataset, we further select high-quality for supervised fine-tuning, ensuring enhanced relevance and precision. The model further undergoes large-scale reinforcement learning without additional supervision, emphasizing the enhancement of its reasoning capabilities and explainability. To validate its effectiveness in complex legal applications, we also conduct human evaluations with legal experts. We develop fine-tuned models based on DeepSeek-R1-Distilled versions, available in three dense configurations: 14B, 32B, and 70B.

Problem

• Insufficient legal expertise in existing legal LLMs
• Lack of comprehensive legal training data
• Enhancing reasoning and explainability in legal applications

Innovation

• Compiled millions of legal documents for training
• Used supervised fine-tuning for enhanced precision
• Applied large-scale reinforcement learning for reasoning