You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fork that refines and isolates the Needle-in-a-haystack (NIAH) test
Needle-test evaluation
Models are tested on different depths and context lengths. Retrieval success is binary, checking for the presence of the reference string in predictions.
About
[EMNLP 2024] LongAlign: A Recipe for Long Context Alignment of LLMs