StateBench: Testing how language models track state changes over long documents
We’re introducing StateBench, a benchmark for long-context state tracking—how well language models keep track of information about entities in long documents as they evolve.
Hugging Face Blog7 min read