US Law Question-Answer Pairs for Language Model Training

673 verified question-answer pairs span all major areas of American law, including constitutional, criminal, civil, and intellectual property. Author Gregorgramtoshi created this cleaned corpus for supervised fine-tuning and retrieval-augmented generation tasks. The dataset was last updated in April 2026.

Use Cases

• Supervised fine-tuning of LLMs on legal domain knowledge using the provided question-answer pairs.
• Evaluating LLM-as-a-Judge capabilities by scoring model responses against the verified answers.
• Building RAG systems for legal consultation by retrieving relevant Q&A pairs based on user queries.
• Analyzing coverage and difficulty across legal subdomains like contract, property, and antitrust law through the categorized questions.

Strengths

• Contains 673 high-quality, cleaned, and verified Q&A pairs.
• Covers a wide range of legal areas including constitutional, criminal, civil, and corporate law.
• Received a Judge Score of 9.4/10 from an evaluation by Grok (xAI).

Limitations

• The dataset size of 673 pairs is relatively small for training large models without augmentation.
• No information is provided on the source or methodology for creating the Q&A pairs, limiting reproducibility.
• The temporal coverage of the legal information is unspecified, which may affect relevance for current law.

Provenance

Source

null

Collection Method

null

Time Range

null

Freshness

Last updated in April 2026.

Geography

United States