Intelligent-Internet/II-Search-4B
II-Search-4B is a 4 billion parameter language model developed by Intelligent-Internet, based on Qwen3-4B, and specialized in information seeking, multi-hop reasoning, and web-integrated search. It excels at complex information retrieval, fact verification, and comprehensive report generation, achieving state-of-the-art performance among models of similar size. With a context length of 40960 tokens, it is optimized for research assistance and factual question answering. This model integrates enhanced tool usage for web search and webpage visits, making it highly effective for evidence-based reasoning.
Loading preview...
Overview
II-Search-4B is a 4 billion parameter language model developed by Intelligent-Internet, built upon the Qwen3-4B architecture. It is specifically fine-tuned for advanced information seeking, multi-hop reasoning, and web-integrated search tasks. The model demonstrates strong capabilities in complex information retrieval, fact verification, and generating comprehensive reports, positioning it as a leading model in its size class for these applications.
Key Capabilities
- Enhanced Tool Usage: Integrates web search and webpage visit tools for internet-aware functionality.
- Multi-hop Reasoning: Features sophisticated planning and improved reasoning thought patterns for complex queries.
- Verified Information Retrieval: Cross-checks information for factual accuracy.
- Comprehensive Report Generation: Excels at producing detailed reports for research queries.
- Strong Factual QA Performance: Achieves significant improvements on benchmarks like OpenAI/SimpleQA (91.8%) and Google/Frames (67.5%) compared to other 4B models.
Training Methodology Highlights
The training involved a multi-phase approach:
- Tool Call Ability Stimulation: Distillation from larger models (Qwen3-235B) to establish function calling on multi-hop datasets.
- Reasoning Improvement: Creation of synthetic problems and refinement of reasoning paths.
- Rejection Sampling & Report Generation: Filtering for high-quality reasoning traces and applying STORM-inspired techniques.
- Reinforcement Learning: Training with datasets like dgslibisey/MuSiQue and an in-house search database.
Intended Use Cases
- Information seeking and factual question answering.
- Research assistance and comprehensive report generation.
- Fact verification and evidence-based reasoning.
- Educational and research applications requiring high factual accuracy.