silvermete0r/qwen2.5-nano-function-master

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kPublished:Apr 8, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The silvermete0r/qwen2.5-nano-function-master is a lightweight, 0.5 billion parameter function-calling model fine-tuned from Qwen/Qwen2.5-0.5B-Instruct. Developed by silvermete0r, it is optimized for efficient function-calling tasks in resource-constrained and privacy-sensitive environments. This model excels at generating structured JSON outputs for local function-calling assistants and edge deployments, achieving significant improvements in JSON validity and argument matching compared to its base model.

Loading preview...

Overview

The silvermete0r/qwen2.5-nano-function-master is a specialized 0.5 billion parameter language model, fine-tuned from the Qwen2.5-0.5B-Instruct base model. Its primary purpose is to provide fast and efficient function-calling capabilities, particularly for local and resource-constrained environments. This model is part of the "Fast Nano SLMs" project, focusing on optimizing inference and memory for function-calling tasks.

Key Capabilities & Performance

This model has been fine-tuned on the Salesforce/xlam-function-calling-60k dataset, demonstrating substantial improvements in function-calling accuracy:

  • JSON Validity: Improved from 45.4% to 98.7% after fine-tuning.
  • Function Name Matching: Increased from 15.0% to 93.3%.
  • Argument Key Matching: Rose from 11.8% to 81.0%.
  • Argument Exact Matching: Enhanced from 8.9% to 69.1%.

Performance metrics highlight its efficiency, with an average throughput of 2.06 samples/s and low VRAM usage (Peak VRAM Reserved: 3120 MB).

Intended Use Cases

This model is specifically designed for scenarios requiring reliable and efficient function calling:

  • Local Function-Calling Assistants: Ideal for offline or air-gapped applications.
  • Edge Deployment: Suitable for devices like laptops, Raspberry Pi 5, and Jetson Nano.
  • Privacy-Sensitive Environments: Enables function calling without external API dependencies.
  • Structured JSON Output Generation: Optimized for producing precise JSON function calls.

It is not intended for open-ended conversational use, but rather for its specialized function-calling role.

Finetunes
1 models