How to build the construction of the local retrieval corpus

[qiaodan-cuhk]

Hi,

Thanks for your impressive work on DeepResearch-9K and the R1 framework! It provides a very solid baseline for agentic RL research. The paper mentions a low-cost autonomous pipeline for synthesizing the "Wikipedia-style summaries" used in the local retriever.

However, I couldn't find the specific scripts for this synthesis process in the repository and huggingface. I would like to confirm if the current retrieval scripts in the repo are designed for this specific summary-based corpus or the original Wiki-18 dataset as used in Search-R1. If it's the former, are there any scripts available to build such a corpus?

Thanks for the great contribution!

Best

[MessiahCorleone]

Hello, I have added the data construction pipeline in data/create_data.py. We construct a summary-based corpus by querying an API to generate an abstract for each question, which are then compiled into the final corpus. Thank you for your attention.