ermiaazarkhalili/Llama-3.2-3B-Instruct_Function_Calling_xLAM

Model Overview

ermiaazarkhalili/Llama-3.2-3B-Instruct_Function_Calling_xLAM is a 3.2 billion parameter language model developed by ermiaazarkhalili. It is built upon the meta-llama/Llama-3.2-3B-Instruct base model and has been specifically fine-tuned for function calling. The training utilized Supervised Fine-Tuning (SFT) with LoRA (Low-Rank Adaptation) adapters and 4-bit quantization, making it efficient for deployment.

Key Capabilities

• Function Calling Optimization: Fine-tuned on the Salesforce/xlam-function-calling-60k dataset to enhance its ability to understand and generate function calls.
• Efficient Training & Inference: Employs LoRA with 4-bit NF4 quantization, enabling efficient training and optimized inference, including available GGUF quantizations for CPU/mixed CPU-GPU usage.
• Base Model Performance: Leverages the capabilities of the Llama-3.2-3B-Instruct base model.
• Configurable Training: Details on learning rate, batch size, epochs, and LoRA parameters are provided for reproducibility and further research.

Good For

• Research: Ideal for exploring language model fine-tuning techniques, particularly for function calling.
• Educational Purposes: Suitable for learning about SFT, LoRA, and function calling in LLMs.
• Prototyping Conversational AI: Can be used to develop and test conversational agents that require function calling capabilities.
• Personal Projects: A good choice for personal AI projects where function calling is a core requirement.

Limitations

This model is primarily trained on English data, has a knowledge cutoff limited to its base model, and may exhibit hallucinations. It was fine-tuned with a 2,048 token context length and is not extensively safety-tuned, requiring appropriate guardrails for production use.