HellaSwag

A commonsense natural-language-inference benchmark of multiple-choice sentence completions that are easy for humans but adversarial for models.

Papers using HellaSwag (11)

QUIET: A Multi-Blank Cascaded Story Cloze Benchmark for LLM Creative Generation Capability2026Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection2025 · 1 citesTurning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection2025 · 1 citesTarget-Oriented Pretraining Data Selection via Neuron-Activated Graph2026Making Bias Non-Predictive: Training Robust LLM Reasoning via Reinforcement Learning2026Data-Free Pruning of Self-Attention Layers in LLMs2025On Robustness and Reliability of Benchmark-Based Evaluation of LLMs2025Uncovering Cross-Linguistic Disparities in LLMs using Sparse Autoencoders2025Self-Reasoning Language Models: Unfold Hidden Reasoning Chains with Few Reasoning Catalyst2025More is Less: The Pitfalls of Multi-Model Synthetic Preference Data in DPO Safety Alignment2025Teuken-7B-Base & Teuken-7B-Instruct: Towards European LLMs2024 · 2 cites