seasparks/Llama-3.1-8B-Instruct-LegalCite

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Apr 5, 2025License:llama3.1Architecture:Transformer Cold

seasparks/Llama-3.1-8B-Instruct-LegalCite is an 8 billion parameter instruction-tuned causal language model, built upon Meta's Llama 3.1 architecture with a 32768 token context length. Developed by SeaSparks, this model is specifically fine-tuned for legal citation extraction and direct quoting from legislative documents like EU regulations. It excels at providing grounded, compliance-minded responses by citing short legal sections directly from input text, making it suitable for research and experimental applications in legal tech.

Loading preview...

Llama-3.1-8B-Instruct-LegalCite Overview

This model, developed by SeaSparks, is an 8 billion parameter instruction-tuned variant of Meta's Llama 3.1, specifically fine-tuned for legal citation extraction. It specializes in generating responses by directly quoting short legal sections from provided text, such as EU regulations, ensuring grounded and transparent answers.

Key Capabilities

  • Legal Citation Extraction: Designed to accurately pull and quote specific legal sections in response to questions.
  • Grounded Responses: Ensures answers are directly supported by the input text, crucial for compliance-minded applications.
  • Efficient Fine-tuning: Developed using a 4-bit MLX version for efficient experimentation on Apple Silicon, then converted to float16 for broad compatibility.
  • Performance Improvement: Achieved significant reductions in test loss (2.4 to ~1.1) and perplexity (11.1 to ~2.9) on its evaluation set.

Good for

  • Research and Experimentation: Ideal for exploring how LLMs can support accurate, quote-based question answering in legal contexts.
  • Prototyping Legal Tech: Useful for testing and demoing applications that require precise references from complex policy texts (e.g., GDPR, EU AI Act).
  • Understanding Citation-Style Prompting: A valuable tool for investigating the capabilities of models in generating direct, verifiable citations.

Note: This model is an experimental prototype, not production-ready, and its outputs should always be human-verified in professional settings. It does not provide legal advice.

Finetunes
2 models