Dark-Davies/fusionai

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kPublished:May 15, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

Dark-Davies/fusionai is a 2 billion parameter Small Reasoning Model (SRM) based on the Qwen3-1.7B architecture, fine-tuned for high-logic reasoning tasks. It features native tool calling for web search and calculator functions, utilizing a ReAct framework for agentic behavior. Optimized for CPU inference, this model is designed for rapid deployment and zero-latency logic, making it suitable for PhD-level research tasks on standard hardware.

Loading preview...

Dark-Davies/fusionai: A Small Reasoning Model for Agentic AI

Dark-Davies/fusionai is a 2 billion parameter model built on the Qwen3-1.7B architecture, specifically fine-tuned as a Small Reasoning Model (SRM). It targets the "Goldilocks Zone" for AI deployment, offering sufficient power for complex tasks while maintaining high inference speed on standard CPU environments.

Key Capabilities

  • Native Tool Calling: Directly outputs structured Action calls for web_search and calculator functions.
  • ReAct Framework: Implements a "Thought -> Action -> Observation -> Final Answer" loop for robust agentic behavior.
  • Efficient Deployment: Utilizes 4-bit merged safetensors for instant loading and high-speed CPU inference, requiring only ~1.1GB of memory.
  • Zero-Latency Logic: Optimized to provide responses in under 10 seconds even on a 2-vCPU environment.

Good For

  • Agentic AI Applications: Ideal for scenarios requiring autonomous decision-making and tool use.
  • Complex Reasoning Tasks: Capable of handling PhD-level research tasks.
  • Resource-Constrained Environments: Designed for efficient operation on standard hardware, including Hugging Face's Free Tier CPU.