app.py

import gradio as gr
import os
import torch
from transformers import (
    AutoTokenizer, 
    AutoModelForCausalLM, 
    pipeline,
    AutoProcessor, 
    MusicgenForConditionalGeneration
)
from scipy.io.wavfile import write
import tempfile
from dotenv import load_dotenv
import spaces  # Assumes Hugging Face Spaces library supports `@spaces.GPU`

# Load environment variables (e.g., Hugging Face token)
load_dotenv()
hf_token = os.getenv("HF_TOKEN")

# ---------------------------------------------------------------------
# Load Llama 3 Model with Zero GPU
# ---------------------------------------------------------------------
@spaces.GPU(duration=120)
def load_llama_pipeline_zero_gpu(model_id: str, token: str):
    try:
        tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
        model = AutoModelForCausalLM.from_pretrained(
            model_id,
            use_auth_token=token,
            torch_dtype=torch.float16,
            device_map="auto",  # Automatically handles GPU allocation
            trust_remote_code=True
        )
        return pipeline("text-generation", model=model, tokenizer=tokenizer)
    except Exception as e:
        return str(e)

# ---------------------------------------------------------------------
# Generate Radio Script
# ---------------------------------------------------------------------
def generate_script(user_input: str, pipeline_llama):
    try:
        system_prompt = (
            "You are a top-tier radio imaging producer using Llama 3. "
            "Take the user's concept and craft a short, creative promo script."
        )
        combined_prompt = f"{system_prompt}\nUser concept: {user_input}\nRefined script:"
        result = pipeline_llama(combined_prompt, max_new_tokens=200, do_sample=True, temperature=0.9)
        return result[0]['generated_text'].split("Refined script:")[-1].strip()
    except Exception as e:
        return f"Error generating script: {e}"

# ---------------------------------------------------------------------
# Load MusicGen Model
# ---------------------------------------------------------------------
@spaces.GPU(duration=120)
def load_musicgen_model():
    try:
        model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
        processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
        return model, processor
    except Exception as e:
        return None, str(e)

# ---------------------------------------------------------------------
# Generate Audio
# ---------------------------------------------------------------------
@spaces.GPU(duration=120)
def generate_audio(prompt: str, audio_length: int, mg_model, mg_processor):
    try:
        mg_model.to("cuda")  # Move the model to GPU
        inputs = mg_processor(text=[prompt], padding=True, return_tensors="pt")
        outputs = mg_model.generate(**inputs, max_new_tokens=audio_length)
        mg_model.to("cpu")  # Return the model to CPU

        sr = mg_model.config.audio_encoder.sampling_rate
        audio_data = outputs[0, 0].cpu().numpy()
        normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")

        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_wav:
            write(temp_wav.name, sr, normalized_audio)
            return temp_wav.name
    except Exception as e:
        return f"Error generating audio: {e}"

# ---------------------------------------------------------------------
# Gradio Interface
# ---------------------------------------------------------------------
def radio_imaging_script(user_prompt, llama_model_id):
    # Load Llama 3 Pipeline with Zero GPU
    pipeline_llama = load_llama_pipeline_zero_gpu(llama_model_id, hf_token)
    if isinstance(pipeline_llama, str):
        return pipeline_llama

    # Generate Script
    script = generate_script(user_prompt, pipeline_llama)
    return script

def radio_imaging_audio(script, audio_length):
    # Load MusicGen
    mg_model, mg_processor = load_musicgen_model()
    if isinstance(mg_processor, str):
        return mg_processor

    # Generate Audio
    audio_data = generate_audio(script, audio_length, mg_model, mg_processor)
    return audio_data

# ---------------------------------------------------------------------
# Interface
# ---------------------------------------------------------------------
with gr.Blocks() as demo:
    gr.Markdown("# 🎧 AI Radio Imaging with Llama 3 + MusicGen (Zero GPU)")

    # Script Generation Section
    with gr.Box():
        gr.Markdown("## Step 1: Generate the Promo Script")
        user_prompt = gr.Textbox(label="Enter your promo idea", placeholder="E.g., A 15-second hype jingle for a morning talk show.")
        llama_model_id = gr.Textbox(label="Llama 3 Model ID", value="meta-llama/Meta-Llama-3-70B")
        generate_script_button = gr.Button("Generate Promo Script")
        script_output = gr.Textbox(label="Generated Script", interactive=False)

        generate_script_button.click(
            fn=radio_imaging_script,
            inputs=[user_prompt, llama_model_id],
            outputs=script_output
        )

    # Audio Generation Section
    with gr.Box():
        gr.Markdown("## Step 2: Generate the Sound")
        audio_length = gr.Slider(label="Audio Length (tokens)", minimum=128, maximum=1024, step=64, value=512)
        generate_audio_button = gr.Button("Generate Sound from Script")
        audio_output = gr.Audio(label="Generated Audio", type="filepath")

        generate_audio_button.click(
            fn=radio_imaging_audio,
            inputs=[script_output, audio_length],
            outputs=audio_output
        )

demo.launch(debug=True)