To approach prolonged context prompts properly, versions require strong remember abilities. The 'Needle In a very Haystack' (NIAH) evaluation actions a model's power to correctly recall info from the large corpus of data. We Improved the robustness of this benchmark by making use of one among thirty random needle/concern pairs for every prompt and