itsZyn/ZynDwarf-1.1

ZynDwarf-1.1: A Compact Agent-Oriented Model

ZynDwarf-1.1, developed by Zyn Models / itsZyn, is a 354.5 million parameter language model built upon Liquid AI's LFM2.5-350M. It is specifically designed for efficient local deployment, targeting CPU-oriented hardware and embedded applications. The model's core differentiator is its focus on agent-oriented workflows, enabling it to decide when a tool is useful, emit structured tool calls, and integrate seamlessly into host-controlled agent loops.

Key Capabilities

• Bilingual Support: Proficient in everyday conversation and explanations in both Spanish and English.
• Programming & Debugging: Handles Python, JavaScript, TypeScript, Bash, HTML/CSS, JSON, and SQL tasks, including debugging and code transformation.
• Agentic Behavior: Supports tool selection, tool-call formatting, tool-result continuation, recovery from tool failures, and multi-turn agent interactions, with a native tool-call template (<|tool_call_start|>[ToolName(arg=value)]<|tool_call_end|>).
• Structured Output: Designed to produce structured responses for various applications.
• Local Inference Optimized: Provided in Safetensors, F16 GGUF, and Q4_K_M GGUF formats, with Q4_K_M offering a 68% size reduction and faster CPU throughput for resource-constrained environments.
• Context Length: Features an architectural maximum position length of 128,000 tokens, with a recommended runtime context of 32,768 tokens.

Ideal Use Cases

• Local CPU Deployment: Excellent for applications requiring a small footprint and efficient execution on standard CPUs.
• Autonomous Agents: Suitable for small autonomous or semi-autonomous agents where the host application manages security and tool execution.
• Developer Tools: Can be embedded into local development tools for coding assistance, debugging, and structured output generation.
• Multilingual Assistants: Effective for conversational agents needing to operate in Spanish and English.