StateBench measures how well LMs remember entity changes over long texts
What is StateBench: Testing how language models track state changes over long documents?
StateBench is a benchmark that checks whether language models can track the evolving state of people, objects, or variables across lengthy documents. It includes both simple updates and multi-step chains that test a model's long-range memory.
Who it is for
Researchers building or evaluating long-context models should use it to see where their architecture succeeds or falls short.
How to try it
Visit the StateBench page on Hugging Face: https://huggingface.co/blog/allenai/statebench