Update app.py

app.py

@@ -26,13 +26,13 @@ from huggingface_hub import hf_hub_download
 
 from infer import load_model, eval_model
 from spkr import SpeakerEmbedding
-
+    
 @spaces.GPU
-def dummy_gpu_check():
-    # Just to let Hugging Face Spaces know this app uses GPU
-    return torch.cuda.is_available()
+def spkr_model_init():
+    spkr_model = SpeakerEmbedding(device="cuda")
+    return spkr_model
 
-spkr_model = SpeakerEmbedding(device="cuda")
+spkr_model = spkr_model_init()
 model, tokenizer, tokenizer_voila, model_type = load_model("maitrix-org/Voila-chat", "maitrix-org/Voila-Tokenizer")
 default_ref_file = "examples/character_ref_emb_demo.pkl"
 default_ref_name = "Homer Simpson"
@@ -63,7 +63,8 @@ def delete_directory(request: gr.Request):
 def add_message(history, message):
     history.append({"role": "user", "content": {"path": message}})
     return history, gr.Audio(value=None), gr.Button(interactive=False)
-
+    
+@spaces.GPU
 def call_bot(history, ref_embs, request: gr.Request):
     formated_history = {
         "instruction": instruction,
@@ -87,7 +88,8 @@ def call_bot(history, ref_embs, request: gr.Request):
         history.append({"role": "assistant", "content": {"text": out['text']}})
 
     return history
-
+    
+@spaces.GPU
 def run_tts(text, ref_embs):
     formated_history = {
         "instruction": "",