DynaPT/

pythia-1b-blindspots

	Pythia-1B Blind Spots Dataset




	
	Model Tested

EleutherAI/pythia-1b-deduped A 1B parameter transformer model trained on the Pile. Not instruction-tuned.

This dataset contains examples where the model produces incorrect or unexpected outputs. It highlights blind spots of Pythia-1B, including:

Arithmetic reasoning Logical reasoning Factual errors Unit conversion errors Multi-step reasoning

	Dataset Structure




	

Column Description

input The… See the full description on the dataset page: https://huggingface.co/datasets/DynaPT/pythia-1b-blindspots.