Release HiBench Synthetic Dataset on Hugging Face

[NielsRogge]

Hi @jzzzzh 🤗

I'm Niels from the open-source team at Hugging Face. I came across your work on HiBench, a benchmark for evaluating LLMs' hierarchical structure reasoning capabilities, on the Hugging Face Papers page: https://huggingface.co/papers/2503.00912. Your work looks very promising!

I noticed that HiBench includes a new synthetic hierarchical dataset for task-specific fine-tuning. To maximize its accessibility and impact, we encourage researchers to host their datasets on Hugging Face. This allows users to easily load the dataset using the datasets library:

from datasets import load_dataset
dataset = load_dataset("your-hf-org-or-username/hibench") 

It would greatly benefit the community if you could host the HiBench dataset on Hugging Face. We can assist you with the process and help optimize the dataset card for better discoverability. Would you be interested in uploading your dataset? You can find a guide here: https://huggingface.co/docs/datasets/loading.

Let me know if you have any questions or need assistance!

Kind regards,

Niels
ML Engineer @ Hugging Face 🤗

[jzzzzh]

Hi Niels,
Wow! Thank you for appreciating our work.
We'll verify the dataset and upload it to Hugging Face as soon as possible!
🤗
Best,
zhuohang

[jzzzzh]

We have released our dataset at https://huggingface.co/datasets/zhuohang/HiBench.

[NielsRogge]

Great! I've opened https://huggingface.co/datasets/zhuohang/HiBench/discussions/1 to link it to the paper page.

Feel free to claim the paper page with your HF account and add the Github and/or project page URLs.

[jzzzzh]

Wow! Thanks a lot.