Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

GitHub - snap-stanford/stark: Official Code of "STaRK: Benchmarking LLM Retrieval on Textual and Relational Knowledge Bases"

May 13, 2024 - github.com
The article introduces STaRK, a large-scale semi-structured retrieval benchmark on Textual and Relational Knowledge Bases developed by researchers from Stanford and Amazon. The benchmark is designed to test how effectively large language models (LLMs) can handle complex interplay between textual and relational requirements in queries. It provides three large-scale knowledge bases across different areas and includes natural-sounding and practical queries that mirror real-life scenarios.

The article also provides detailed instructions on how to access the benchmark data, set up the environment, load the data, and evaluate the benchmark. It also mentions the release of STaRK SKB Explorer on Hugging Face, an interactive interface for exploring their knowledge bases. The STaRK benchmark was presented at the 2024 Stanford Annual Affiliates Meeting and 2024 Stanford Data Science Conference.

Key takeaways:

  • STaRK is a large-scale semi-structure retrieval benchmark on Textual and Relational Knowledge Bases, designed to extract nodes from the knowledge base that are relevant to a user query.
  • The benchmark includes three large-scale knowledge bases across different areas, constructed from public sources.
  • The queries in the benchmark are designed to incorporate rich relational information and complex textual properties, and closely mirror questions in real-life scenarios.
  • The STaRK benchmark was presented at the 2024 Stanford Annual Affiliates Meeting and 2024 Stanford Data Science Conference.
View Full Article

Comments (0)

Be the first to comment!