Integrate BrowseComp

[cwing-nvidia]

Background

BrowseComp (Browser-based Composition) is a benchmark for evaluating agents on web search and information gathering tasks. Users have requested integration of this benchmark in NeMo Gym

What is BrowseComp

BrowseComp evaluates agents on their ability to navigate the web, search for information, and synthesize answers from multiple sources.

Acceptance Criteria

• [ ] Implement BrowseComp in NeMo Gym