Nanthasit/sakthai-context-0.5b-tools

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 4, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Nanthasit/sakthai-context-0.5b-tools is an ultra-light 0.5 billion parameter Qwen2.5-0.5B fine-tuned model by Nanthasit, specifically optimized for reliable tool/function calling in conversational agents. It achieves 91.2% selection accuracy on SakThai Bench v2 with 0% degenerate outputs. This model is designed for edge deployment, running efficiently on consumer hardware with approximately 1 GB RAM, making it ideal for on-device assistants and Raspberry Pi applications.

Loading preview...

SakThai Context 0.5B Tools: Ultra-Light Tool-Calling Agent

SakThai Context 0.5B Tools is a specialized 0.5 billion parameter model developed by Nanthasit, built upon the Qwen2.5-0.5B-Instruct architecture. It is a prompt-masked supervised fine-tune (SFT) specifically engineered for robust tool and function calling within conversational AI agents. The model's primary focus is on accurately selecting the correct tool, generating valid JSON-style arguments, and preventing degenerate outputs, making it highly reliable for automated interactions.

Key Capabilities & Features

  • Optimized Tool Calling: Achieves 91.2% selection accuracy on SakThai Bench v2, demonstrating strong performance in identifying and utilizing tools.
  • Reliable Output: Boasts 0% degenerate outputs in multi-trial evaluations, ensuring consistent and usable responses.
  • Ultra-Lightweight: Designed for efficiency, it can run on consumer hardware with approximately 1 GB of RAM, making it suitable for resource-constrained environments.
  • Base Model: Fine-tuned from Qwen/Qwen2.5-0.5B-Instruct.
  • Max Position: Supports a maximum context length of 32768 tokens.

Ideal Use Cases

  • Lightweight Agents: Perfect for building efficient and responsive AI agents.
  • On-Device Assistants: Enables AI functionalities directly on user devices.
  • Edge Deployments: Suitable for applications on hardware like Raspberry Pi, where computational resources are limited.
  • Browser/Tool Calling: Specifically trained for scenarios requiring interaction with external tools or browser environments.

Limitations

Due to its compact 0.5B parameter scale, the model has limitations in reasoning depth, resulting in lower arguments accuracy compared to selection accuracy. Adherence to complex tool schemas and long context traces may also degrade performance.

Finetunes
2 models