ermiaazarkhalili/Qwen3-8B-Function-Calling-xLAM-Unsloth

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Apr 25, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The ermiaazarkhalili/Qwen3-8B-Function-Calling-xLAM-Unsloth model is an 8 billion parameter Qwen3-based causal language model, fine-tuned by ermiaazarkhalili for function calling. It leverages Unsloth for optimized training, resulting in 2x faster fine-tuning and 60% less VRAM usage. This model excels at interpreting user queries and tool definitions to generate structured function call responses, making it suitable for applications requiring robust tool-use capabilities.

Loading preview...

Overview

This model, developed by ermiaazarkhalili, is an 8 billion parameter Qwen3-based language model specifically fine-tuned for function calling. It utilizes the Unsloth framework for efficient training, achieving 2x faster training speeds and 60% reduced VRAM consumption compared to standard methods. The model was trained on the Salesforce/xlam-function-calling-60k dataset, comprising 60,000 examples of queries, tool definitions, and structured answers.

Key Capabilities

  • Function Calling: Optimized to understand natural language requests and generate appropriate function calls based on provided tool definitions.
  • Efficient Training: Fine-tuned using Unsloth with QLoRA (4-bit) for resource-efficient adaptation.
  • Qwen3 Base: Built upon the Qwen3-8B architecture, offering a strong foundation for language understanding.
  • GGUF Availability: Provides quantized GGUF versions for CPU and edge inference, supporting various deployment scenarios.

Good For

  • Tool Use Applications: Ideal for integrating with external tools and APIs where structured function calls are required.
  • Agentic Workflows: Enhancing AI agents with the ability to interact with their environment through defined functions.
  • Resource-Constrained Environments: The Unsloth optimization makes it suitable for fine-tuning and deployment with limited GPU resources.

Limitations

  • Primarily trained on English data.
  • Context length is limited to 2,048 tokens during fine-tuning.
  • May exhibit hallucinations and is not extensively safety-tuned.

Finetunes
2 models