app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import spaces

# Dictionary to store loaded models and tokenizers
loaded_models = {}

def load_model(model_name):
    """Load the model and tokenizer if not already loaded."""
    if model_name not in loaded_models:
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(
            model_name, torch_dtype=torch.float16, device_map="auto"
        )
        loaded_models[model_name] = (tokenizer, model)
    return loaded_models[model_name]

@spaces.GPU
def generate_text(model_name, prompt):
    """Generate text using the selected model."""
    tokenizer, model = load_model(model_name)
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=256)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# List of models to choose from
model_choices = [
    "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
    "meta-llama/Llama-3.2-3B-Instruct",
    "google/gemma-7b"
]

# Gradio interface setup
with gr.Blocks() as demo:
    gr.Markdown("## Clinical Text Analysis with Multiple Models")
    model_selector = gr.Dropdown(choices=model_choices, label="Select Model")
    input_text = gr.Textbox(label="Input Clinical Text")
    output_text = gr.Textbox(label="Generated Output")
    analyze_button = gr.Button("Analyze")

    analyze_button.click(fn=generate_text, inputs=[model_selector, input_text], outputs=output_text)

demo.launch()