analiza aiqbench
This research addresses query redundancy in agentic search by introducing DivInit, a training-free method that selects diverse initial queries from multiple candidates to improve parallel sampling efficiency. The approach yields consistent performance gains across open-weight models and benchmarks without increasing computational depth or token usage per trajectory.
Beyond Parallel Sampling: Diverse Query Initialization for Agentic Search
arXiv:2606.17209v1 Announce Type: new Abstract: Test-time scaling for agentic search typically increases depth (i.e., more turns and tokens per trajectory) or breadth (i.e., more parallel rollouts). Here we focus on breadth scaling, showing that standard parallel sampling yields diminishing returns, tracing this to query redundancy at the first turn. When models issue similar first queries across rollouts, the threads retrieve overlapping evidence, and subsequent turns are conditioned on this shared retrieval. We address this limitation with DivInit, a training-free intervention at the first turn. Rather than sampling k independent first queries, DivInit draws n candidates from a single call, picks k < n diverse seeds, and runs them as parallel trajectories. Across five open-weight models and eight benchmarks, DivInit consistently improves over standard parallel sampling, with average gains of five to seven points on multi-hop QA at matched compute. Code available at https://github.com/cxcscmu/diverse-query-initialization
Abstract:Test-time scaling for agentic search typically increases depth (i.e., more turns and tokens per trajectory) or breadth (i.e., more parallel rollouts). Here we focus on breadth scaling, showing that standard parallel sampling yields diminishing returns, tracing this to query redundancy at the first turn. When models issue similar first queries across rollouts, the threads retrieve overlapping evidence, and subsequent turns are conditioned on this shared retrieval. We address this limitation with DivInit, a training-free intervention at the first turn. Rather than sampling k independent first queries, DivInit draws n candidates from a single call, picks k < n diverse seeds, and runs them as parallel trajectories. Across five open-weight models and eight benchmarks, DivInit consistently improves over standard parallel sampling, with average gains of five to seven points on multi-hop QA at matched compute. Code available at this https URL
Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
arXiv:2606.17209 [cs.AI]
(or arXiv:2606.17209v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.17209 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 237 cuvinte
Subiecte conexe
Initialization Improves LLM-Driven Discovery
arXiv · 2 oct. 2026
Supabase Pipelines is now in Public Alpha
Supabase · 21 iul. 2026
Parallelism, critical windows, and separations among diffusion language models
arXiv · 18 sept. 2026
Deep recurrent models are less robustly CoT-monitorable than normal CoT models in a toy setting
LessWrong · 18 sept. 2026