silvermete0r/qwen2.5-nano-function-master
silvermete0r/qwen2.5-nano-function-master is a 0.5 billion parameter language model fine-tuned from Qwen/Qwen2.5-0.5B-Instruct, specifically optimized for function-calling tasks. It leverages the Salesforce/xlam-function-calling-60k dataset to achieve high accuracy in generating valid JSON function calls. This model is designed for efficient inference in resource-constrained or private environments, making it ideal for local function-calling assistants and edge deployments.
Loading preview...
Overview
silvermete0r/qwen2.5-nano-function-master is a compact 0.5 billion parameter model, part of the Fast Nano SLMs project, engineered for efficient function-calling. It is fine-tuned from Qwen/Qwen2.5-0.5B-Instruct using the Salesforce/xlam-function-calling-60k dataset, focusing on optimizing inference and memory for specialized function-calling tasks.
Key Capabilities
- High Function-Calling Accuracy: Achieves significant improvements in JSON validity (98.7%), function name matching (93.3%), and argument parsing (81.0% keys, 69.1% exact) after fine-tuning.
- Resource-Efficient: Designed for low-resource environments, offering fast inference with an average of 154 tokens per second and minimal VRAM usage (peak 3120 MB).
- Structured Output: Excels at generating precise JSON arrays of function calls based on user prompts and provided tool definitions.
Good For
- Local & Offline Assistants: Ideal for function-calling applications that need to run without internet access or API calls.
- Edge Deployment: Suitable for devices like laptops, Raspberry Pi 5, and Jetson Nano due to its small size and efficiency.
- Privacy-Sensitive Use Cases: Ensures data remains on-device, making it perfect for applications requiring strict privacy.
- Structured JSON Generation: Specifically tuned for reliable output of structured JSON for tool use.
Not Intended For
- Open-ended conversational AI or general-purpose chat.