initiaize on cpu for frontend model

cosyvoice/cli/frontend.py

@@ -37,7 +37,8 @@ class CosyVoiceFrontEnd:
                  allowed_special: str = 'all'):
         self.tokenizer = get_tokenizer()
         self.feat_extractor = feat_extractor
-        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+        #self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+        self.device = 'cpu'
         option = onnxruntime.SessionOptions()
         option.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
         option.intra_op_num_threads = 1
@@ -109,12 +110,14 @@ class CosyVoiceFrontEnd:
         return texts
 
     def frontend_sft(self, tts_text, spk_id):
+        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
         tts_text_token, tts_text_token_len = self._extract_text_token(tts_text)
         embedding = self.spk2info[spk_id]['embedding']
         model_input = {'text': tts_text_token, 'text_len': tts_text_token_len, 'llm_embedding': embedding, 'flow_embedding': embedding}
         return model_input
 
     def frontend_zero_shot(self, tts_text, prompt_text, prompt_speech_16k):
+        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
         tts_text_token, tts_text_token_len = self._extract_text_token(tts_text)
         prompt_text_token, prompt_text_token_len = self._extract_text_token(prompt_text)
         prompt_speech_22050 = torchaudio.transforms.Resample(orig_freq=16000, new_freq=22050)(prompt_speech_16k)