Albert-C-Ong/lexis-phi4-obligation-generator

TEXT GENERATIONConcurrency Cost:1Model Size:14.7BQuant:FP8Ctx Length:32kPublished:Mar 22, 2026License:cc-by-4.0Architecture:Transformer Open Weights Cold

The Albert-C-Ong/lexis-phi4-obligation-generator is a 14.7 billion parameter model, QLoRA fine-tuned from Microsoft's Phi-4 architecture. It specializes in extracting legal obligations from contract text and formatting them into structured JSON. This model is optimized for legal tech applications requiring precise identification of contractual duties, ownership, deadlines, and risk levels. Its primary strength lies in generating machine-readable legal obligation data from unstructured text.

Loading preview...

LEXIS Phi-4 Obligation Generator Overview

This model, developed by Albert-C-Ong, is a 14.7 billion parameter language model based on Microsoft's Phi-4 architecture. It has been specifically fine-tuned using QLoRA on the CUAD dataset to excel at identifying and extracting legal obligations from contract documents. The fine-tuning process involved 4-bit NF4 quantization, LoRA with r=64 and alpha=128, a learning rate of 0.0002, and a cosine scheduler over 3 epochs, with a maximum sequence length of 2048 tokens.

Key Capabilities

  • Legal Obligation Extraction: Precisely identifies contractual obligations within legal texts.
  • Structured JSON Output: Formats extracted information into a machine-readable JSON structure, including fields like action, deontic_type, owner, deadline, and risk_level.
  • Specialized Legal AI: Tailored for legal tech applications requiring automated analysis of contracts.

Good For

  • Automating the identification of legal obligations in contracts.
  • Generating structured data from unstructured legal documents.
  • Applications requiring the extraction of specific legal clauses and their attributes (e.g., deadlines, responsible parties).

Finetunes
1 models