Update app.py

app.py

@@ -7,50 +7,70 @@ import scipy.io.wavfile
 import io
 import numpy as np
 
-# 1. Model Loading
+# 1. Resource-Optimized Model Loading
 MODEL_ID = "facebook/musicgen-small"
 device = "cpu"
 
-print(f"Loading model {MODEL_ID}...")
+print(f"DEBUG: Loading model {MODEL_ID}...")
 processor = AutoProcessor.from_pretrained(MODEL_ID)
 model = MusicgenForConditionalGeneration.from_pretrained(MODEL_ID, torch_dtype=torch.float32)
 model.to(device)
-print("Model loaded successfully.")
+print("DEBUG: Model loaded and ready on CPU.")
 
 def generate_core(prompt, duration):
     if not prompt:
+        print("DEBUG: Request received with no prompt.")
         return None
     
-    duration = min(int(duration), 30)
-    inputs = processor(text=[prompt], padding=True, return_tensors="pt").to(device)
-    max_tokens = int(duration * 50)
+    print(f"DEBUG: Starting generation for: '{prompt}' ({duration}s)")
     
-    with torch.no_grad():
-        audio_values = model.generate(**inputs, max_new_tokens=max_tokens)
+    try:
+        duration = min(int(duration), 30)
+        inputs = processor(text=[prompt], padding=True, return_tensors="pt").to(device)
+        
+        # 50 tokens = 1 second of audio
+        max_tokens = int(duration * 50)
+        
+        with torch.no_grad():
+            audio_values = model.generate(
+                **inputs, 
+                max_new_tokens=max_tokens,
+                do_sample=True,
+                guidance_scale=3.0
+            )
 
-    sampling_rate = model.config.audio_encoder.sampling_rate
-    audio_data = audio_values[0, 0].cpu().numpy()
-    return sampling_rate, audio_data
+        sampling_rate = model.config.audio_encoder.sampling_rate
+        audio_data = audio_values[0, 0].cpu().numpy()
+        
+        print("DEBUG: Generation successful.")
+        return sampling_rate, audio_data
+    except Exception as e:
+        print(f"DEBUG ERROR: {str(e)}")
+        return None
 
-# 2. FastAPI Setup
+# 2. FastAPI Engine
 app = FastAPI()
 
 @app.post("/generate")
 async def api_generate(prompt: str = Query(...), duration: int = Query(10)):
     try:
-        sr, audio = generate_core(prompt, duration)
+        res = generate_core(prompt, duration)
+        if res is None:
+            raise HTTPException(status_code=400, detail="Generation failed or prompt empty")
+        
+        sr, audio = res
         byte_io = io.BytesIO()
         scipy.io.wavfile.write(byte_io, rate=sr, data=audio)
         return Response(content=byte_io.getvalue(), media_type="audio/wav")
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
-# 3. Gradio Interface
+# 3. Gradio Interface with Queue enabled
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 🎵 MusicGen Automation Hub")
     with gr.Row():
         with gr.Column():
-            p_in = gr.Textbox(label="Prompt", placeholder="Upbeat synthwave...")
+            p_in = gr.Textbox(label="Prompt", placeholder="Enter music description...")
             d_in = gr.Slider(1, 30, value=10, label="Duration (sec)")
             run_btn = gr.Button("Generate", variant="primary")
         with gr.Column():
@@ -58,7 +78,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     
     run_btn.click(generate_core, [p_in, d_in], a_out)
 
-# 4. Mounting without the incompatible 'show_api' argument
-# We mount it at the root. The API documentation crash is bypassed by 
-# the explicit Query parameters in the FastAPI routes above.
+# 4. Mandatory: Enable Queuing
+# This prevents the UI from "hanging" during the 60-120s CPU generation time.
+demo.queue()
+
+# Mount Gradio
 app = gr.mount_gradio_app(app, demo, path="/")