Decision Boundary for Search Agent - Search-R1 DPO v3 overled data
This dataset belongs to the Decision Boundary for Search Agent work and is separated from other reasonrag projects by the decision-boundary-search-agent naming prefix and tag.
It contains 6200 DPO preference pairs for post-training Search-R1 style search agents. The data is organized to reduce over-search while retaining under-search fixes.
Approximate composition: