Parsee Datasets: Structured Information Extraction Benchmarks for MLLMs

Parsee Datasets provides benchmarks and case studies for extracting structured information from PDFs, HTML files, and images, released by the Parsee.ai team. Updated in May 2024, the collection is designed to evaluate and train Multimodal Large Language Models (MLLMs) for document parsing tasks. The repository includes specialized data for Retrieval-Augmented Generation (RAG) workflows.

Use Cases

• Benchmarking MLLM performance on structured data extraction from PDF files
• Evaluating RAG pipelines using the provided document-based case studies
• Training models to convert HTML content into structured JSON or tabular formats

Strengths

• Includes multiple source formats including PDF, HTML, and images
• MIT licensed for open commercial and research use
• Specifically tailored for MLLM and RAG benchmarking

Limitations

• Lack of documented record counts or schema details in the primary repository
• Narrow focus on extraction tasks rather than general-purpose natural language processing

Provenance

Source

parsee-ai

Collection Method

Curated and annotated by the Parsee.ai team

Freshness

Last updated May 2024.