thangvip/qwen3-1.7b-vietnamese-legal-grpo-phase-2
The thangvip/qwen3-1.7b-vietnamese-legal-grpo-phase-2 is a 1.7 billion parameter Qwen 3 model developed by thangvip, specifically fine-tuned for advanced Vietnamese legal reasoning. This Phase 2 model utilizes Group Relative Policy Optimization (GRPO) on hard difficulty legal question-answering data, excelling at complex syllogistic reasoning and generating structured legal arguments with citations. It supports an extended context length of up to 8192 tokens, making it suitable for intricate legal analysis.
Loading preview...
Overview
This model, thangvip/qwen3-1.7b-vietnamese-legal-grpo-phase-2, is a Phase 2 Vietnamese legal reasoning specialist built on the Qwen 3 (1.7B parameters) architecture. It is fine-tuned using Group Relative Policy Optimization (GRPO) on hard difficulty Vietnamese legal question-answering data, focusing on complex syllogistic reasoning (Major Premise → Minor Premise → Conclusion).
Key Capabilities
- Advanced Legal Reasoning: Specifically designed for challenging, multi-step legal scenarios in Vietnamese.
- Syllogistic Structure: Generates responses formatted with XML-like tags for clear, structured legal arguments.
- Citation Support: Includes relevant legal citations in its output.
- Extended Context: Supports up to 8192 tokens, enabling analysis of longer and more complex legal texts.
- Two-Phase Training: Builds upon a Phase 1 model trained on normal difficulty questions, progressively mastering harder problems.
Use Cases
- Complex Legal Education: Ideal for teaching advanced legal reasoning methodologies.
- Advanced Legal Research: Provides preliminary analysis for intricate legal questions.
- Multi-step Legal Analysis: Generates structured arguments for complex scenarios.
- Legal Training: Demonstrates proper syllogistic reasoning for difficult legal problems.