Abdelrahman350/

smollm3-3b-blindspots

	SmolLM3-3B Blind Spot Dataset




	
	Tested Model

HuggingFaceTB/SmolLM3-3B — a raw pretrained base LLM (no instruction tuning) released July 8 2025 with 3B parameters.

Loaded on a free Colab T4 GPU in 4-bit NF4 quantisation with bitsandbytes: from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig import torch

MODEL_ID = "HuggingFaceTB/SmolLM3-3B"

bnb_config = BitsAndBytesConfig( load_in_4bit=True… See the full description on the dataset page: https://huggingface.co/datasets/Abdelrahman350/smollm3-3b-blindspots.