Am

The 'Am' dataset/benchmark is used to evaluate input-centric uncertainty quantification methods for large language models by providing a structured way to analyze and attribute uncertainty contributions from specific spans of input.

Papers using Am (2)

Localizing Input Uncertainty Quantification for Large Language Models via Shapley Values2026RecRankerEval: A Flexible and Extensible Framework for Top-k LLM-based Recommendation2025