NimahMasuud/

model_mistakes

	Nanbeige4-3B-Base — Blind Spots Dataset




	
	1. Model Tested

Model: https://huggingface.co/Nanbeige/Nanbeige4-3B-Base This is a 3-billion parameter base language model released in December 2025 by Nanbeige LLM Lab. It is a pre-trained only model, not fine-tuned for any specific application.

The model was loaded in Google Colab using a free T4 GPU. !pip install transformers torch accelerate -q

from transformers import… See the full description on the dataset page: https://huggingface.co/datasets/NimahMasuud/model_mistakes.