app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import login
import os
import json

# Authenticate with HuggingFace using environment variable
hf_token = os.getenv('HF_TOKEN')
if hf_token:
    login(token=hf_token)

# Load FunctionGemma model and tokenizer
model_name = "google/functiongemma-270m-it"
tokenizer = AutoTokenizer.from_pretrained(model_name, token=hf_token)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float32,
    device_map="auto",
    token=hf_token
)

# Define example tools for function calling
tools = [
    {
        "type": "function",
        "function": {
            "name": "calculate",
            "description": "Performs mathematical calculations",
            "parameters": {
                "type": "object",
                "properties": {
                    "operation": {
                        "type": "string",
                        "description": "The mathematical operation (add, subtract, multiply, divide)"
                    },
                    "a": {
                        "type": "number",
                        "description": "First number"
                    },
                    "b": {
                        "type": "number",
                        "description": "Second number"
                    }
                },
                "required": ["operation", "a", "b"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "send_email",
            "description": "Sends an email to a specified recipient",
            "parameters": {
                "type": "object",
                "properties": {
                    "to": {
                        "type": "string",
                        "description": "Email recipient address"
                    },
                    "subject": {
                        "type": "string",
                        "description": "Email subject"
                    },
                    "body": {
                        "type": "string",
                        "description": "Email body content"
                    }
                },
                "required": ["to", "subject", "body"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "set_alarm",
            "description": "Sets an alarm for a specified time",
            "parameters": {
                "type": "object",
                "properties": {
                    "time": {
                        "type": "string",
                        "description": "Time in HH:MM format (24-hour)"
                    },
                    "label": {
                        "type": "string",
                        "description": "Label or description for the alarm"
                    }
                },
                "required": ["time"]
            }
        }
    }
]

def test_function_calling(user_input: str, temperature: float = 0.7) -> str:
    """Test FunctionGemma model for function calling."""
    try:
        # Prepare messages with DEVELOPER role for function calling activation
        messages = [
            {
                "role": "developer",
                "content": "You are a helpful assistant that can make function calls. When the user asks you to do something that matches one of the available functions, call that function."
            },
            {
                "role": "user",
                "content": user_input
            }
        ]
        
        # Format input with tools for the model
        formatted_input = tokenizer.apply_chat_template(
            messages,
            tools=tools,
            tokenize=False,
            add_generation_prompt=True
        )
        
        # Tokenize
        inputs = tokenizer(formatted_input, return_tensors="pt").to(model.device)
        
        # Generate
        with torch.no_grad():
            outputs = model.generate(
                **inputs,
                max_new_tokens=512,
                temperature=temperature,
                top_p=0.95,
                do_sample=True
            )
        
        # Decode output
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return result
    except Exception as e:
        return f"Error: {str(e)}"

# Create Gradio interface
with gr.Blocks(title="FunctionGemma-270M Function Calling Tester") as demo:
    gr.Markdown("# FunctionGemma-270M Function Calling Tester")
    gr.Markdown("Test the FunctionGemma model's ability to generate function calls.")
    
    with gr.Row():
        with gr.Column():
            user_input = gr.Textbox(
                label="Input",
                placeholder="Describe a function call you want",
                lines=3
            )
            temperature = gr.Slider(
                minimum=0.1,
                maximum=1.0,
                value=0.7,
                label="Temperature"
            )
            submit_btn = gr.Button("Test Model", variant="primary")
        
        with gr.Column():
            output = gr.Textbox(
                label="Model Output",
                lines=10
            )
    
    # Example prompts
    gr.Examples(
        examples=[
            ["Calculate 5 + 3"],
            ["Send email to user@example.com"],
            ["Set alarm for 6 AM"]
        ],
        inputs=[user_input]
    )
    
    submit_btn.click(test_function_calling, inputs=[user_input, temperature], outputs=output)

demo.launch()