Needle-In-A-Haystack (NIAH)

The Needle-In-A-Haystack (NIAH) dataset/benchmark contains retrieval tasks designed to evaluate the long-context recall capabilities of hybrid linear-attention models.

Papers using Needle-In-A-Haystack (NIAH) (4)

NoLiMa: Long-Context Evaluation Beyond Literal Matching2025 · 4 citesAttention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix It2026Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache2025Infinite Retrieval: Attention Enhanced LLMs in Long-Context Processing2025