Datasets:

---

internlm

/

STAR-Bench

like 3

Follow

Intern Large Models 838

This pull request significantly enhances the STAR-Bench dataset card by:

• Adding task_categories: ['audio-text-to-text'] to the metadata, reflecting the dataset's focus on audio reasoning yielding textual responses.
• Including language: ['en'] as the primary language for the dataset's textual components and documentation.
• Introducing descriptive tags such as 4d-intelligence, spatio-temporal-reasoning, audio-reasoning, and audio-benchmark to improve discoverability.
• Updating the header links to explicitly include both the Hugging Face paper page (https://huggingface.co/papers/2510.24693) and the arXiv preprint link (https://arxiv.org/abs/2510.24693), replacing the previous empty arXiv placeholder.
• Adding a comprehensive "Sample Usage" section, extracted and adapted from the GitHub repository's "Test Your Model!" guide, to help users easily set up and evaluate models with the dataset. The huggingface-cli download command in this section has been corrected to use internlm/STAR-Bench for direct download.
• Replacing the "TBD" placeholder in the "Citation" section with the full BibTeX entry provided in the GitHub README.
• Including the "Acknowledgement" section from the GitHub README to credit platform support.

These changes collectively make the dataset card more informative, discoverable, and user-friendly.