bqbbao6/Qwen2.5-3B-legal-vn

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kPublished:May 26, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

bqbbao6/Qwen2.5-3B-legal-vn is a 3.1 billion parameter Qwen2.5-Instruct model, fine-tuned by bqbbao6 and optimized with Unsloth for Vietnamese legal domain mastery. This 4-bit quantized model excels at context-based question answering (RAG) by strictly adhering to provided legal texts, drastically reducing hallucination. It is designed for high-fidelity legal consulting within the Vietnamese administrative and legal sectors, offering hardware efficiency for deployment on mid-range GPUs.

Loading preview...

bqbbao6/Qwen2.5-3B-legal-vn: Vietnamese Legal Domain LLM

This model is a specialized fine-tuned version of Qwen2.5-3B-Instruct, developed by bqbbao6. It leverages Unsloth for accelerated training and is quantized to 4-bit (bitsandbytes), making it highly efficient for deployment. The primary focus of this model is Vietnamese Legal Domain Mastery and Context-Based Question Answering (RAG).

Key Capabilities:

  • Strict Context Adherence: Engineered to prioritize and reason directly over provided legal articles, decrees, and circulars, minimizing hallucination in legal consulting.
  • Legal Formalism: Adopts the precise, formal, and authoritative tone essential for the Vietnamese administrative and legal sectors.
  • Hardware Efficiency: Operates with a low VRAM footprint, enabling long contexts and tool-calling structures on mid-range GPUs.
  • Fine-tuning Method: Utilizes QLoRA (Rank 16, Alpha 32) for efficient adaptation.

Good For:

  • Retrieval-Augmented Generation (RAG) pipelines requiring high-fidelity legal information.
  • Applications needing a specialized legal agent for the Vietnamese context.
  • Deployments where hardware efficiency and reduced VRAM usage are critical.