alibaba-pai/SearchQwen2.5-3B
SearchQwen2.5-3B is a 3.09 billion parameter Search Agent model developed by Alibaba Cloud PAI, built upon the Qwen2.5-3B-Instruct base. It is specifically trained with environment-aligned, solver-verified search trajectories using EasyDistill 2.0. This model excels at multi-hop search, browsing, and evidence integration, making it highly effective for structured tool-call interactions in search-related tasks.
Loading preview...
SearchQwen2.5-3B: A Compact Search Agent
SearchQwen2.5-3B is a specialized 3.09 billion parameter model from Alibaba Cloud PAI, designed as a compact Search Agent. It is built on the Qwen2.5-3B-Instruct base and features a 32,768 token context length.
Key Capabilities & Training
This model is uniquely trained using EasyDistill 2.0 with environment-aligned, solver-verified search trajectories from the SynSearch-Data dataset. Its primary strength lies in its ability to handle structured search and browse tool calls, making it highly effective for complex information retrieval tasks.
Performance Highlights
SearchQwen2.5-3B demonstrates significant performance improvements over its base model in search-related benchmarks. For Tool-Call interactions, it achieves an overall accuracy of 35.00%, compared to 21.60% for Qwen2.5-3B-Instruct. Specifically, it scores 48.58% on Multi-hop QA and 21.40% on Deep Search tasks, indicating its proficiency in multi-hop reasoning and integrating information from various sources.
Ideal Use Cases
This model is particularly well-suited for applications requiring:
- Multi-hop search: Answering complex questions that require multiple search queries.
- Web browsing and evidence integration: Systematically gathering and synthesizing information from web sources.
- Structured tool-calling: Acting as an agent that can effectively use external search and browse tools to fulfill user requests.