Datasets:

---

AceSearcher

/

evaluation_datasets

like 1

Follow

AceSearcher 4

This PR significantly improves the dataset card for AceSearcher by:

• Adding task_categories: ['question-answering', 'text-retrieval', 'text-generation'] to the metadata to improve discoverability.
• Including relevant tags like rag, llm, reasoning, search, multi-hop-reasoning, fact-verification, and reinforced-self-play.
• Setting the language to en.
• Providing a clear introduction to AceSearcher, linking to the paper (AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforced Self-Play) and the associated GitHub repository (https://github.com/ritaranx/AceSearcher/).
• Listing direct links to the various SFT, RFT, and evaluation datasets hosted on Hugging Face in a dedicated "Data Download" table.
• Including "Data Generation", "Evaluation", and "Training" sections from the GitHub README to give users context on the dataset's lifecycle and related processes.
• Adding a comprehensive "Sample Usage" section with all available code snippets directly from the GitHub README, demonstrating how to perform question decomposition, sub-question answering, fact verification, and final answer generation for various tasks, including document-level financial reasoning.
• Including the BibTeX citation.

These updates provide a comprehensive and informative resource for researchers interested in AceSearcher and its associated datasets.