Create app.py

app.py

@@ -0,0 +1,36 @@
+import gradio as gr
+import tempfile
+import torch
+import scipy.io.wavfile as wavfile
+from transformers import AutoProcessor, MusicgenForConditionalGeneration
+
+# Load model
+processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
+model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model.to(device)
+
+def generate_music(prompt, duration):
+    if not prompt.strip():
+        return None, "Please enter a prompt."
+    inputs = processor(text=[prompt], return_tensors="pt").to(device)
+    max_new_tokens = int(256 * (duration / 8))  # simple scaling
+    with torch.no_grad():
+        audio = model.generate(**inputs, max_new_tokens=max_new_tokens)
+    sr = model.config.audio_encoder.sampling_rate
+    audio_arr = audio[0, 0].cpu().numpy()
+    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+    wavfile.write(tmp.name, rate=sr, data=audio_arr)
+    return tmp.name, f"Generated {duration}s of audio!"
+
+with gr.Blocks(title="MusicGen 🎶") as demo:
+    gr.Markdown("# 🎵 MusicGen — Text-to-Music Generator (Small Model)")
+    with gr.Row():
+        prompt = gr.Textbox(label="Describe your music", placeholder="e.g. lo-fi hip hop with piano")
+        duration = gr.Slider(4, 20, value=8, step=1, label="Duration (seconds)")
+    btn = gr.Button("Generate 🎧")
+    audio_out = gr.Audio(label="Output", type="filepath")
+    msg = gr.Textbox(label="Status", interactive=False)
+    btn.click(generate_music, inputs=[prompt, duration], outputs=[audio_out, msg])
+
+demo.launch()