FahmidAfrin/

nanbeige-3b-blind-spots

	Nanbeige-3B Blind Spots Dataset




	
	Model Tested

We evaluated the base language model Nanbeige/Nanbeige4-3B-Base (3B parameters), available at:https://huggingface.co/Nanbeige/Nanbeige4-3B-Base
This is a pretrained base model (not instruction-tuned), making it suitable for studying inherent model blind spots.

We tested the model across diverse task categories including factual QA, translation, arithmetic, commonsense reasoning, negation… See the full description on the dataset page: https://huggingface.co/datasets/FahmidAfrin/nanbeige-3b-blind-spots.