Update app.py

app.py

@@ -23,18 +23,18 @@ auth_token = os.environ.get("AUTH_TOKEN") or True
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 print(f"Using device: {device}")
 
-@spaces.GPU(duration=60)
+@spaces.GPU(duration=60 * 2)
 def pipe(file, return_timestamps=False):
-    model = WhisperForConditionalGeneration.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, low_cpu_mem_usage=True)
-    model.to(device)
-    processor = WhisperProcessor.from_pretrained(MODEL_NAME)
-    model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
-    model.generation_config.cache_implementation = "static"
+    # model = WhisperForConditionalGeneration.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, low_cpu_mem_usage=True)
+    # model.to(device)
+    # processor = WhisperProcessor.from_pretrained(MODEL_NAME)
+    # model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
+    # model.generation_config.cache_implementation = "static"
     asr = pipeline(
         task="automatic-speech-recognition",
-        model=model,
-        tokenizer=AutoTokenizer.from_pretrained(MODEL_NAME),
-        feature_extractor=AutoFeatureExtractor.from_pretrained(MODEL_NAME),
+        model=MODEL_NAME,
+        # tokenizer=AutoTokenizer.from_pretrained(MODEL_NAME),
+        # feature_extractor=AutoFeatureExtractor.from_pretrained(MODEL_NAME),
         chunk_length_s=30,
         device=device,
         token=auth_token,