Legal Dialogue Scenarios Reformatted for Instruction Tuning

Lawyer-Instruct contains conversational legal data reshaped into instruction, input, and expected output format, designed for supervised dialogue model training. The dataset was created by Alignment-Lab-AI and released on the Hugging Face platform in July 2023. It is a reformatted version of the original LawyerChat dataset.

Use Cases

• Fine-tune a language model to generate legal advice by training on the 'instruction' and 'expected output' fields.
• Train a model to respond to user queries ('input') within legal scenarios using the structured dialogue examples.
• Benchmark model performance on legal reasoning tasks by evaluating outputs against the provided 'expected output' field.

Strengths

• Dataset is formatted specifically for instruction-following model training with three key fields.
• Contains at least 10,000 data points as indicated by the platform tag 'Size Categories1 Kn10 K'.

Limitations

• Specific column details, total row count, and original data sources are not provided in the available description.
• The dataset is primarily in English, limiting applicability for multilingual legal dialogue models.

Provenance

Source

Reformatted from the original LawyerChat dataset.

Collection Method

Legal dialogue scenarios were reshaped into an instruction-tuning format.

Freshness

Last updated on 2023-07-14.

Geography

Region tag suggests a US focus, but specific coverage is unknown.