dolev31/ProactiveInquirer-Qwen3-8B-Merged

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 27, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The dolev31/ProactiveInquirer-Qwen3-8B-Merged is an 8 billion parameter Qwen3-based model, developed by Ido Levy, Asaf Yehudai, Segev Shlomov, Asaf Adi, and Leshem Choshen from IBM and Weizmann Institute of Science. This model is specifically fine-tuned as a "questioner" agent, designed to proactively ask clarifying questions to gather necessary information for complex tasks, rather than directly answering. It operates by generating JSON actions (ask or stop) based on a prompt template, making it suitable for integration into agentic workflows requiring intelligent information retrieval.

Loading preview...

ProactiveInquirer-Qwen3-8B-Merged Overview

This model is a merged version of the "questioner" agent developed by Ido Levy et al. from IBM and Weizmann Institute of Science, based on the Qwen3-8B architecture. It integrates a LoRA adapter to create a standard full-weight model, loading without PEFT and serving like any other Qwen3-8B variant. The primary innovation lies in its ability to proactively inquire for information, a concept detailed in the paper "Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents" (arXiv:2609.37236).

Key Capabilities

  • Proactive Questioning: Designed to act as a "questioner" within an agent system, it generates clarifying questions to gather missing information. This contrasts with traditional LLMs that aim to answer directly.
  • JSON Action Output: Replies with structured JSON actions, either {"action": "ask", "question": ...} or {"action": "stop", ...}, facilitating integration into automated workflows.
  • Agent Component: Intended as a specialized component within a larger agent architecture, rather than a standalone chat assistant.
  • Standard Deployment: Functions as a full-weight Qwen3-8B model, compatible with vLLM, SGLang, or TGI for deployment.

Good For

  • Building Advanced AI Agents: Ideal for developers creating agents that need to intelligently seek out information before attempting to solve complex problems.
  • Information Retrieval Systems: Enhancing systems where initial queries may be ambiguous or incomplete, requiring follow-up questions to refine understanding.
  • Automated Task Execution: Use cases where an agent needs to break down a task into smaller, information-gathering steps.
  • Research in Agent Proactivity: A valuable tool for researchers exploring horizontal and vertical proactivity in AI agents.

Finetunes
1 models