metadata
tags:
- gguf
- llama.cpp
- unsloth
- mind_call
- function_call
datasets:
- frshafi/mind_call
language:
- en
base_model:
- meta-llama/Llama-3.2-3B-Instruct
metrics:
- accuracy: 88%
health_function_call_llama3.2_3b_gguf: GGUF
A fine-tuned Llama 3.2 3B GGUF model designed for structured function calling in healthcare edge devices.Trained to convert natural language health queries into JSON-based function calls.
Base Model: LLama 3.2 3B
Fine Tuning: Parameter Efficient Fine Tuning. Targeted all linear layers (Q, K, V, O, gate, up, down), the model learned complex mapping logic while maintaining a tiny 10.5 MB adapter footprint.
Quantization: Exported to GGUF (Q4_K_M) format.
Dataset: The model is trained on the MindCall Dataset, a curated synthetic collection of 5,000+ high-fidelity health interaction pairs.
🚀 Key Features
- Converts user queries → structured API calls
- Lightweight GGUF format (runs locally via llama.cpp)
- Optimized for deterministic outputs (low temperature)
- Supports reasoning via
<think>tags
📦 Model Files
Llama-3.2-3B-Instruct.Q4_K_M.gguf
⚡ Quick Start (Python)
Install dependencies
pip install llama-cpp-python huggingface_hub
Load the model
from llama_cpp import Llama
llm = Llama.from_pretrained(
repo_id="ramgovindv/health_function_call_llama3.2_3b_gguf",
filename="Llama-3.2-3B-Instruct.Q4_K_M.gguf",
)
Inference
query = "I am feeling dizzy for 2 days"
prompt = f"""
You are an API generator.
Return JSON in this format:
{{
"name": "function_name",
"parameters": {{
"key": "value"
}}
}}
User query:
{query}
JSON:
"""
response = llm.create_chat_completion(
messages=[{"role": "user", "content": prompt}],
temperature=0.1
)
output = response["choices"][0]["message"]["content"]
print(output)
Output
<think>
User has dizziness → likely need blood pressure check
</think>
<function>
{
"name": "get_blood_pressure_data",
"parameters": {
"num_days": 2
}
}
</function>
<think> → reasoning
<function> → actual function call
