Solutions for standardizing the format of legal instruments
WritingBench solves the challenge of standardizing the format of legal documents in the following ways:
- Structured training:: Selection of tasks such as contract clauses, indictments, etc. under the category "Legal", which on average contain 5-7 standardized format modules
- Specialized Judging Models: The downloadable Qwen-7B judging model has been pre-trained to recognize elements such as numbering systems, citation formats, etc., specific to legal instruments.
- Dynamic detection mechanisms: Modify format scoring weights in criter.py (default accounts for 20%) to focus on checking:
- Article numbering continuity
- Arrangement of party information
- Space reserved for signature area
- Remediation program: Based on the formatting deductions in scores.jsonl, add the corresponding formatting instruction template to prompt.py
Implementation point: It is recommended to run short text tasks (e.g., legal statements) under the benchmark_query/requirement/length category before gradually expanding to full instrument writing.
This answer comes from the articleWritingBench: a benchmarking assessment tool to test the writing skills of large modelsThe
May not be reproduced without permission:AI productivity tools " How to overcome the formatting errors that large models are prone to in legal writing?
