RekklesAI/Qwen2.5-Coder-32B-Glaive-ToolCall
RekklesAI/Qwen2.5-Coder-32B-Glaive-ToolCall is a 32.8 billion parameter large language model, fine-tuned from Qwen/Qwen2.5-Coder-32B-Instruct. This transformer-based decoder model is specifically enhanced for tool calling capabilities, leveraging the Glaive Function Calling v2 dataset. It excels at understanding function schemas, context-aware tool selection, and robust JSON generation for API integration and automation workflows. The model is designed for building AI assistants and intelligent automation that interact with external systems.
Loading preview...
What is RekklesAI/Qwen2.5-Coder-32B-Glaive-ToolCall?
This model is a 32.8 billion parameter Large Language Model (LLM) developed by RekklesAI, fine-tuned from the Qwen/Qwen2.5-Coder-32B-Instruct base model. Its core differentiator is its significantly enhanced tool calling capabilities, achieved through fine-tuning with the Glaive Function Calling v2 dataset.
Key Capabilities & Enhancements
- Superior Tool Calling: Demonstrates improved understanding of complex function schemas, context-aware tool selection, and accurate parameter extraction from natural language.
- Robust JSON Generation: Produces well-formatted JSON for function calls, adhering to proper schema.
- Multi-step Tool Orchestration: Enhanced ability to chain multiple tool calls for complex tasks.
- API Integration: Improved understanding of various web service interfaces like REST APIs and GraphQL.
- Code Generation with Tools: Can generate code that effectively incorporates external tool usage.
Training Details
The model was fine-tuned using LoRA (Low-Rank Adaptation) over 3 epochs on 100,000 samples from the Glaive Function Calling v2 dataset. This dataset, comprising 113,000 training examples, focuses on diverse and realistic function calling scenarios, including single and multi-step workflows, error handling, and parameter validation.
When to Use This Model
This model is particularly well-suited for applications requiring intelligent interaction with external systems and APIs:
- AI Assistants: Building conversational agents that can execute actions via tools.
- Automation Workflows: Creating dynamic scripts that leverage external services.
- Code Generation: Generating code snippets that integrate with APIs.
- System Integration: Bridging different software systems through tool-based interactions.
Limitations
Its performance is primarily optimized for patterns found in the Glaive Function Calling v2 dataset, meaning performance may vary for highly specialized tools not represented in the training data. Careful prompt engineering is also crucial for optimal tool calling performance.