Update app.py

app.py

@@ -31,16 +31,26 @@ class AudioRequest(BaseModel):
     prompt: str
 
 # --- Model Loading ---
-# This section runs once when the application starts.
 device = "cuda" if torch.cuda.is_available() else "cpu"
-torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+torch_dtype = torch.float16 if "cuda" in device else torch.float32
 
 logger.info(f"Using device: {device} with dtype: {torch_dtype}")
 
+# --- FIX FOR PERMISSION ERROR ---
+# The environment we're running in doesn't allow writing to the default '/.cache' directory.
+# We explicitly define a writable directory within '/tmp' for the model cache.
+CACHE_DIR = "/tmp/huggingface_cache"
+os.makedirs(CACHE_DIR, exist_ok=True)
+logger.info(f"Using model cache directory: {CACHE_DIR}")
+
 try:
     # Use the stable, recommended model
     repo_id = "cvssp/audioldm-s-full-v2"
-    pipe = AudioLDMPipeline.from_pretrained(repo_id, torch_dtype=torch_dtype)
+    pipe = AudioLDMPipeline.from_pretrained(
+        repo_id, 
+        torch_dtype=torch_dtype,
+        cache_dir=CACHE_DIR # Pass the writable cache directory to the loader
+    )
     pipe = pipe.to(device)
     logger.info(f"Successfully loaded model: {repo_id}")
 except Exception as e:
@@ -60,7 +70,6 @@ async def generate_audio_endpoint(request: AudioRequest):
     temp_file_path = ""
     try:
         # Generate the audio waveform
-        # These are good parameters for this model
         audio = pipe(
             prompt, 
             num_inference_steps=200, 
@@ -68,9 +77,6 @@ async def generate_audio_endpoint(request: AudioRequest):
             guidance_scale=7.0
         ).audios[0]
 
-        # --- THIS IS THE FIX ---
-        # The cvssp/audioldm-s-full-v2 model has a fixed sample rate of 16000 Hz.
-        # We set it directly here instead of trying to read it from a config that no longer exists.
         sample_rate = 16000
 
         # Create a temporary file to save the audio
@@ -85,7 +91,6 @@ async def generate_audio_endpoint(request: AudioRequest):
             logger.info(f"Audio saved to temporary file: {temp_file_path}")
 
         # Return the audio file as a response. 
-        # The file will be deleted after being sent.
         return FileResponse(
             path=temp_file_path, 
             media_type='audio/wav',