This PR updates the dataset card for the ROME benchmark with the following changes:
- Corrected `task_categories` in the metadata from `visual-question-answering` to `image-text-to-text` to better reflect the visual reasoning nature of the benchmark.
- Added relevant `tags` (`benchmark`, `reasoning`, `vlm`) to the metadata for improved discoverability.
- Updated the "Home & Leaderboard" link in the header to point to the official project page (`https://flageval-baai.github.io/LRM-Eval/`).
- Updated the "Paper" link in the header and the introductory paper link to `https://huggingface.co/papers/2509.17177`. The introductory paper link text was also updated to reflect the full paper title from the bibtex entry.
- Updated the "Data" link in the header to point to the specific Hugging Face dataset repository (`https://huggingface.co/datasets/FlagEval/ROME`).
These changes aim to make the dataset card more accurate and informative.
