autolabel

Autolabel is a Python library that automates the labeling, cleaning, and enrichment of text datasets by routing them through large language models via a JSON configuration file. Users define a task type (classification, named entity recognition, question answering, and others), specify labeling guidelines and few-shot examples, select a model provider, then call a LabelingAgent to run the labeling pipeline against a CSV dataset. It connects to Claude through Anthropic's API, supporting models such as claude-3-opus-20240229 alongside OpenAI, Google, and HuggingFace-hosted models. A built-in plan step previews the final prompt and estimates cost before any labels are generated, a practical safeguard for large datasets. The library also includes a benchmarking suite that runs identical prompts across all supported models and outputs a results.csv for direct comparison. The primary audience is machine learning engineers and data scientists who need annotated training data at lower cost and faster turnaround than manual labeling workflows.

What people ask about autolabel