updated

app.py

@@ -18,7 +18,7 @@ MODEL_CONFIG = {
         'savings_dollars': 8.4,  # $8.4 saved per 1000 hours
         'savings_percent': 74.9,  # 74.9% savings
         'compressed_cost': 2.8,  # $2.8 for compressed
-        'original_cost': 11.3,   # $11.3 for original
+        'original_cost': 11.3,  # $11.3 for original
     },
     'batch_mode': True,
     'batch_size': 2  # Number of variants to generate (2, 4, 6, etc.)
@@ -49,13 +49,22 @@ def cleanup_temp_files():
     import time
     temp_dir = tempfile.gettempdir()
     cutoff_time = time.time() - 3600
-    for temp_file in glob.glob(os.path.join(temp_dir, "tmp*.wav")):
-        try:
-            if os.path.getctime(temp_file) < cutoff_time:
-                os.remove(temp_file)
-                print(f"[CLEANUP] Removed old temp file: {temp_file}")
-        except OSError:
-            pass
+    
+    # Clean old generated music files
+    patterns = [
+        os.path.join(temp_dir, "tmp*.wav"),
+        os.path.join(temp_dir, "generated_music_*.wav"),
+        os.path.join(temp_dir, "musicgen_variant_*.wav")
+    ]
+    
+    for pattern in patterns:
+        for temp_file in glob.glob(pattern):
+            try:
+                if os.path.getctime(temp_file) < cutoff_time:
+                    os.remove(temp_file)
+                    print(f"[CLEANUP] Removed old temp file: {temp_file}")
+            except OSError:
+                pass
 
 
 _generator = None
@@ -317,6 +326,9 @@ def generate_music_batch(text_prompt, duration=10, guidance_scale=3.0, model_mod
         audio_variants = []
         sample_rate = outputs[0]['sampling_rate']
         
+        # Create unique timestamp for this generation batch
+        batch_timestamp = int(time.time() * 1000)
+        
         for i, output in enumerate(outputs):
             audio_data = output['audio']
             
@@ -343,7 +355,17 @@ def generate_music_batch(text_prompt, duration=10, guidance_scale=3.0, model_mod
                 audio_data = audio_data / max_val * 0.95
             
             audio_data = (audio_data * 32767).astype(np.int16)
-            audio_variants.append((sample_rate, audio_data))
+            
+            # Save each variant to a unique temporary file
+            temp_filename = f"musicgen_variant_{i+1}_{batch_timestamp}.wav"
+            temp_path = os.path.join(tempfile.gettempdir(), temp_filename)
+            
+            sf.write(temp_path, audio_data, sample_rate)
+            
+            print(f"[GENERATION] Variant {i + 1} saved to: {temp_path}")
+            print(f"[GENERATION] Variant {i + 1} file size: {os.path.getsize(temp_path)} bytes")
+            
+            audio_variants.append(temp_path)
             
             print(f"[GENERATION] Variant {i + 1} final shape: {audio_data.shape}")
 
@@ -355,7 +377,8 @@ def generate_music_batch(text_prompt, duration=10, guidance_scale=3.0, model_mod
         variants_text = "audio"
         generation_info = f"✅ Generated {variants_text} in {generation_time:.2f}s\n{savings_message}"
         
-        return audio_variants[0], audio_variants[1], audio_variants[2], audio_variants[3], audio_variants[4], audio_variants[5], generation_info
+        return audio_variants[0], audio_variants[1], audio_variants[2], audio_variants[3], audio_variants[4], \
+               audio_variants[5], generation_info
 
     except Exception as e:
         print(f"[ERROR] Batch generation failed: {str(e)}")
@@ -366,11 +389,11 @@ def generate_music_batch(text_prompt, duration=10, guidance_scale=3.0, model_mod
 
 with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
     gr.Markdown("# 🎵 MusicGen Large Music Generator")
-    
+
     gr.Markdown(
         f"Generate music from text descriptions using Facebook's MusicGen "
         f"Large model accelerated by TheStage for 2.3x faster performance.")
-    
+
     with gr.Column():
         text_input = gr.Textbox(
             label="Music Description",
@@ -378,7 +401,7 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
             lines=3,
             value="A groovy funk bassline with a tight drum beat"
         )
-        
+
         with gr.Row():
             duration = gr.Slider(
                 minimum=5,
@@ -395,35 +418,35 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
                 label="Guidance Scale",
                 info="Higher values follow prompt more closely"
             )
-        
+
         generate_btn = gr.Button("🎵 Generate Music", variant="primary", size="lg")
-        
+
         generation_info = gr.Markdown("Ready to generate music with elastic acceleration")
-        
+
         audio_section_title = "### Generated Music"
         gr.Markdown(audio_section_title)
-        
+
         actual_outputs = MODEL_CONFIG['batch_size'] if MODEL_CONFIG['batch_mode'] else 1
-        
+
         audio_outputs = []
-        
+
         with gr.Row():
-            audio_output1 = gr.Audio(label="Variant 1", type="numpy", visible=actual_outputs >= 1)
-            audio_output2 = gr.Audio(label="Variant 2", type="numpy", visible=actual_outputs >= 2)
+            audio_output1 = gr.Audio(label="Variant 1", type="filepath", visible=actual_outputs >= 1)
+            audio_output2 = gr.Audio(label="Variant 2", type="filepath", visible=actual_outputs >= 2)
             audio_outputs.extend([audio_output1, audio_output2])
         
         with gr.Row():
-            audio_output3 = gr.Audio(label="Variant 3", type="numpy", visible=actual_outputs >= 3)
-            audio_output4 = gr.Audio(label="Variant 4", type="numpy", visible=actual_outputs >= 4)
+            audio_output3 = gr.Audio(label="Variant 3", type="filepath", visible=actual_outputs >= 3)
+            audio_output4 = gr.Audio(label="Variant 4", type="filepath", visible=actual_outputs >= 4)
             audio_outputs.extend([audio_output3, audio_output4])
         
         with gr.Row():
-            audio_output5 = gr.Audio(label="Variant 5", type="numpy", visible=actual_outputs >= 5)
-            audio_output6 = gr.Audio(label="Variant 6", type="numpy", visible=actual_outputs >= 6)
+            audio_output5 = gr.Audio(label="Variant 5", type="filepath", visible=actual_outputs >= 5)
+            audio_output6 = gr.Audio(label="Variant 6", type="filepath", visible=actual_outputs >= 6)
             audio_outputs.extend([audio_output5, audio_output6])
-        
+
         savings_banner = gr.Markdown(get_fixed_savings_message())
-        
+
         with gr.Accordion("💡 Tips & Information", open=False):
             gr.Markdown(f"""
             **Generation Tips:**
@@ -431,19 +454,22 @@ with gr.Blocks(title="MusicGen Large - Music Generation") as demo:
             - Higher guidance scale = follows prompt more closely
             - Lower guidance scale = more creative/varied results
             - Duration is limited to 30 seconds for faster generation
-            
+
             **Performance:**
             - Accelerated by TheStage elastic compression
             - L40S GPU pricing: $1.8/hour
             """)
 
+
     def generate_simple(text_prompt, duration, guidance_scale):
         return generate_music_batch(text_prompt, duration, guidance_scale, "compressed")
 
+
     generate_btn.click(
         fn=generate_simple,
         inputs=[text_input, duration, guidance_scale],
-        outputs=[audio_output1, audio_output2, audio_output3, audio_output4, audio_output5, audio_output6, generation_info],
+        outputs=[audio_output1, audio_output2, audio_output3, audio_output4, audio_output5, audio_output6,
+                 generation_info],
         show_progress=True
     )