serteal/sparse-probing
Sparse Probing Datasets 155 binary classification tasks for probing language model representations. From: "Are Sparse Autoencoders Useful? A Case Study in Sparse Probing" (arXiv:2502.16681) Source: EleutherAI/sae-probes Usage from datasets import load_dataset # Load a specific dataset ds = load_dataset("serteal/sparse-probing", "87_glue_cola") # List available configurations from datasets import get_dataset_config_names configs =… See the full description on the dataset page: https://huggingface.co/datasets/serteal/sparse-probing.
Sparse Probing Datasets
155 binary classification tasks for probing language model representations.
From: "Are Sparse Autoencoders Useful? A Case Study in Sparse Probing" (arXiv:2502.16681)
Source: EleutherAI/sae-probes
Usage
from datasets import load_dataset
# Load a specific dataset
ds = load_dataset("serteal/sparse-probing", "87_glue_cola")
# List available configurations
from datasets import get_dataset_config_names
configs = get_dataset_config_names("serteal/sparse-probing")
print(f"Available: {len(configs)} datasets")Dataset Categories
Available Datasets
Knowledge
Reasoning
Science
Moderation
Glue
Sentiment
Other
Code
Schema
Each dataset contains:
prompt(string): The input textlabel(int): Binary label (0 or 1)
Citation
@article{makelov2025sparse,
title={Are Sparse Autoencoders Useful? A Case Study in Sparse Probing},
author={Makelov, Aleksandar and others},
journal={arXiv preprint arXiv:2502.16681},
year={2025}
}License
Apache 2.0