use cpu instead of gpu for testing

app.py

@@ -42,9 +42,9 @@ EXAMPLES_DIR = "audio"
 EXAMPLE_FILE = os.path.join(EXAMPLES_DIR, "PromoterClipMono.wav")
 
 # Chunk processing settings
-DEFAULT_CHUNK_DURATION = 30  # seconds per chunk
-OVERLAP_DURATION = 2  # seconds of overlap between chunks
-MAX_DURATION_WITHOUT_CHUNKING = 60  # auto-chunk if longer than this
+DEFAULT_CHUNK_DURATION = 5  # seconds per chunk
+OVERLAP_DURATION = 1  # seconds of overlap between chunks
+MAX_DURATION_WITHOUT_CHUNKING = 10  # auto-chunk if longer than this
 
 # Global model cache
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
@@ -153,7 +153,7 @@ def save_audio(tensor, sample_rate):
         return tmp.name
 
 
-@spaces.GPU(duration=300)
+#@spaces.GPU(duration=300)
 def separate_audio(model_name, file_path, text_prompt, chunk_duration=DEFAULT_CHUNK_DURATION, progress=gr.Progress()):
     global model, processor