Update app.py

app.py

@@ -35,17 +35,19 @@ translate_token_id = all_special_ids[-6]
 
 def _preload_and_load_models():
     global vicuna_tokenizer, vicuna_model
-    VICUNA_MODEL_NAME = "EleutherAI/gpt-neo-2.7B"  # Or another model
+    #VICUNA_MODEL_NAME = "EleutherAI/gpt-neo-2.7B"  # Or another model
+    #VICUNA_MODEL_NAME = "lmsys/vicuna-13b-v1.5"  # Or another model
+    VICUNA_MODEL_NAME = "lmsys/vicuna-7b-v1.5"  # Or another model
     vicuna_tokenizer = AutoTokenizer.from_pretrained(VICUNA_MODEL_NAME)
     vicuna_model = AutoModelForCausalLM.from_pretrained(
         VICUNA_MODEL_NAME,
         torch_dtype=torch.float16,
-        device_map="auto", # or.to('cuda')
-    ) #.to('cuda') # Explicitly move to CUDA after loading
+     #   device_map="auto", # or.to('cuda')
+    ).to('cuda') # Explicitly move to CUDA after loading
 
 _preload_and_load_models()
 
-tts = Text2Speech.from_pretrained("espnet/kan-bayashi_ljspeech_vits")
+tts = Text2Speech.from_pretrained("espnet/kan-bayashi_ljspeech_vits").to('cuda')
 
 @spaces.GPU(required=True)
 def process_audio(microphone, state, task="transcribe"):