BoyBarley/BoyBarley-Sparky-v3

TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kPublished:Apr 24, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

BoyBarley Sparky v3 is a 0.5 billion parameter autonomous AI assistant developed by BoyBarley, based on Qwen2.5-0.5B-Instruct, with a 32768 token context length. It is fine-tuned for coding, server management, and task automation, excelling in tool-native JSON output and safety-first refusal. This bilingual model (Indonesian and English) is designed for lightweight operation on CPU/VM environments, offering fast inference and consistent identity.

Loading preview...

BoyBarley Sparky v3: The Fast, Professional, Energetic AI Assistant

BoyBarley Sparky v3, or "Barley," is a highly efficient 0.5 billion parameter autonomous AI assistant built on the Qwen2.5-0.5B-Instruct base model. It is specifically fine-tuned for coding, server management, and task automation, emphasizing a safety-first approach. This model is notable for its lightweight footprint, capable of running on CPU/VMs with 1GB RAM (Q4 version), and its fast inference speeds of 50+ tokens/second on modern CPUs.

Key Capabilities & Differentiators

  • Tool-Native: Generates valid JSON tool calls for 8 standardized tools, including server, read, write, exec, browser, cron, nodes, and message.
  • Safety by Design: Consistently refuses destructive commands (sudo, rm -rf) and adheres to strict sandbox constraints for file operations and shell execution.
  • Bilingual Proficiency: Fine-tuned with a curated Indonesian and English dataset, making it effective in both languages.
  • Consistent Identity: Maintains a grounded identity as "Barley," avoiding confusion with its base model.
  • Strong Benchmarks: Achieves an "EXCELLENT" overall score of 89.92% on the Autonomous Assistant Benchmark, with high scores in Safety Refusal (95.58%), General Q&A (100%), and Code Generation (88.88%).

Ideal Use Cases

  • Automated Server Management: Executing commands like checking disk usage, restarting services, and viewing logs through structured tool calls.
  • Code Generation & Assistance: Generating Python functions and other code snippets efficiently.
  • Task Automation: Handling repetitive tasks with its tool-calling capabilities.
  • Resource-Constrained Environments: Its small size and CPU-friendly nature make it suitable for deployment on VMs or devices with limited RAM.

Finetunes
3 models