vtgh1602/legal-llm-sft-v4-qwen25-7b-merged

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:May 9, 2026Architecture:Transformer Cold

The vtgh1602/legal-llm-sft-v4-qwen25-7b-merged model is a 7.6 billion parameter language model based on the Qwen2.5 architecture, specifically fine-tuned for legal applications. This model contains the merged 16-bit weights for the SFT v4 legal adapter, designed to work in conjunction with a grounded retriever like vtgh1602/legal-llm-faiss-v2. Its primary purpose is to process and generate legal-specific text, requiring external citation verification for substantive answers and indicating insufficient information if retrieval is weak.

Loading preview...

Overview

This model, vtgh1602/legal-llm-sft-v4-qwen25-7b-merged, is a 7.6 billion parameter language model built upon the Qwen2.5 architecture. It incorporates the merged 16-bit weights from the SFT v4 legal adapter, making it specialized for legal domain tasks. The model is intended to be used as part of a larger system, specifically paired with a grounded retriever.

Key Capabilities

  • Legal Domain Specialization: Fine-tuned with a legal adapter (SFT v4) to understand and generate legal-specific text.
  • Integration with Retrievers: Designed to work in tandem with external retrieval systems, such as vtgh1602/legal-llm-faiss-v2, to provide grounded responses.
  • Citation Requirement: Emphasizes the necessity of citation verification for any substantive legal answers it provides, ensuring accuracy and reliability.
  • Handling Weak Retrieval: Includes a mechanism to indicate when there is insufficient information (Chua du can cu de ket luan) if the associated retrieval process yields weak or empty results.

Good For

  • Legal Information Processing: Ideal for applications requiring the processing and generation of text within the legal domain.
  • Grounded Legal Q&A Systems: Suitable as a core component in systems that aim to provide legally sound answers by leveraging external knowledge bases.
  • Downstream Quantization and Deployment: The merged 16-bit weights are prepared for further quantization and deployment in various legal tech solutions.

Finetunes
1 models