app.py

import gradio as gr
import sambanova_gradio

# Define a function to load the selected model
def load_model(model_choice):
    model = gr.load(
        name=model_choice,
        src=sambanova_gradio.registry
    )
    return model

# Available model choices
model_choices = [
    "Meta-Llama-3.2-1B-lnstruct",
    "Meta-Llama-3.2-3B-Instruct",
    "Meta-Llama-3.1-88-Instruct",
    "Meta-Llama-3.1-405B-lnstruct",
    "Meta-Llama-3.1-70B-lnstruct"
]

# Create Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Meta-Llama Model Selector")
    
    # Dropdown for selecting the model
    model_choice = gr.Dropdown(
        choices=model_choices, 
        label="Choose a Meta-Llama Model",
        value="Meta-Llama-3.2-3B-Instruct"  # default value
    )
    
    # Textbox for user input and output
    input_box = gr.Textbox(
        label="Input",
        placeholder="Ask a question...",
        examples=["Explain quantum gravity to a 5-year old.", 
                  "How many R are there in the word Strawberry?"]
    )
    
    # Output for the model's response
    output_box = gr.Textbox(label="Output")
    
    # Button to generate the output
    generate_button = gr.Button("Generate Response")
    
    # Action when button is clicked
    def generate_response(model_choice, user_input):
        model = load_model(model_choice)
        output = model.predict(user_input)
        return output
    
    generate_button.click(
        generate_response, 
        inputs=[model_choice, input_box], 
        outputs=output_box
    )

# Launch the Gradio app
demo.launch(share=True)