alibaba-pai/SearchQwen2.5-7B
SearchQwen2.5-7B is a 7.62 billion parameter Search Agent model developed by Alibaba Cloud PAI, built upon the Qwen2.5-7B-Instruct base. It is specifically trained with environment-aligned, solver-verified search trajectories using EasyDistill 2.0. This model excels in multi-hop search, browsing, and evidence integration, making it ideal for applications requiring structured tool-call for web interaction.
Loading preview...
SearchQwen2.5-7B: A Specialized Search Agent Model
SearchQwen2.5-7B, developed by Alibaba Cloud PAI, is a 7.62 billion parameter model designed as a compact Search Agent. It is built on the Qwen/Qwen2.5-7B-Instruct base and leverages EasyDistill 2.0 for training, utilizing environment-aligned, solver-verified search trajectories.
Key Capabilities
- Multi-hop Search: Optimized for complex search queries requiring multiple steps.
- Browsing and Evidence Integration: Capable of navigating web content and synthesizing information.
- Structured Tool-Call: Designed to interact with external search and browse tools using a structured interface.
- Enhanced Performance: Outperforms its base model, Qwen2.5-7B-Instruct, in both Multi-hop QA and Deep Search benchmarks, particularly when using structured tool-calls.
Training and Data
The model was trained using the SynSearch-Data dataset, focusing on generating effective search and browse actions. It supports a context length of 32,768 tokens.
When to Use This Model
SearchQwen2.5-7B is particularly well-suited for applications that require an intelligent agent to perform web-based information retrieval, answer complex multi-hop questions, and integrate findings from various sources. Its strength lies in its ability to generate structured tool calls for search and browse operations, making it an excellent choice for building sophisticated search-augmented systems.