evaluation-task · Source-linked discovery

Needle in a Haystack (NIAH): In-Context Retrieval Benchmark for Long Context LLMs

NIAH evaluates in-context retrieval ability of long context LLMs by testing a model's ability to extract factual information from long-context inputs.

Open interactive record →
OriginElliot Nelson, Georgios Kollias, Payel Das et al.TopicsGeneral capabilityStatusimported

Can support

Not independently assessed by FronteraEval yet.

Cannot support by itself

No inference beyond the upstream source should be made until the protocol is reviewed.

Original sources