Update app.py

app.py

@@ -48,7 +48,7 @@ import gradio as gr
 #if not torch.cuda.is_available():
     #DESCRIPTION += "\n<p>⚠️Running on CPU, This may not work on CPU.</p>"
 
-CACHE_EXAMPLES = torch.device('cuda') and os.getenv("CACHE_EXAMPLES", "0") == "1"
+#CACHE_EXAMPLES = torch.device('cuda') and os.getenv("CACHE_EXAMPLES", "0") == "1"
 #CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "0") == "1"
 #USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
 #ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
@@ -58,16 +58,18 @@ device = torch.device('cuda')
 
 def transcribe(file_upload, progress=gr.Progress(track_tqdm=True)): # microphone
 
-    file = file_upload # microphone if microphone is not None else
+    file = file_upload 
     start_time = time.time()
 
     #--------------____________________________________________--------------"
-
     with torch.no_grad():
-        pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", chunk_length_s=30, device=device)
+        pipe = pipeline("automatic-speech-recognition", 
+                        model="NbAiLab/nb-whisper-large", 
+                        chunk_length_s=30, 
+                        forced_decoder_ids=None,  # -- explicitly set to None
+                        device=device)
 
     text = pipe(file)["text"]
-
     #--------------____________________________________________--------------"
 
     end_time = time.time()