Remove flow_cache arg to match CosyVoice3 flow.inference

cosyvoice/cli/model.py

@@ -58,7 +58,7 @@ class CosyVoiceModel:
         self.tts_speech_token_dict = {}
         self.llm_end_dict = {}
         self.mel_overlap_dict = {}
-        self.flow_cache_dict = {}
+        # self.flow_cache_dict = {}
         self.hift_cache_dict = {}
         self.silent_tokens = []
 
@@ -134,14 +134,14 @@ class CosyVoiceModel:
 
     def token2wav(self, token, prompt_token, prompt_feat, embedding, uuid, finalize=False, speed=1.0):
         with torch.cuda.amp.autocast(self.fp16):
-            tts_mel, self.flow_cache_dict[uuid] = self.flow.inference(token=token.to(self.device, dtype=torch.int32),
+            tts_mel, _ = self.flow.inference(token=token.to(self.device, dtype=torch.int32),
                                                                       token_len=torch.tensor([token.shape[1]], dtype=torch.int32).to(self.device),
                                                                       prompt_token=prompt_token.to(self.device),
                                                                       prompt_token_len=torch.tensor([prompt_token.shape[1]], dtype=torch.int32).to(self.device),
                                                                       prompt_feat=prompt_feat.to(self.device),
                                                                       prompt_feat_len=torch.tensor([prompt_feat.shape[1]], dtype=torch.int32).to(self.device),
                                                                       embedding=embedding.to(self.device),
-                                                                      flow_cache=self.flow_cache_dict[uuid])
+                                                                      )
 
         # mel overlap fade in out
         if self.mel_overlap_dict[uuid].shape[2] != 0:
@@ -183,7 +183,7 @@ class CosyVoiceModel:
             self.tts_speech_token_dict[this_uuid], self.llm_end_dict[this_uuid] = [], False
             self.hift_cache_dict[this_uuid] = None
             self.mel_overlap_dict[this_uuid] = torch.zeros(1, 80, 0)
-            self.flow_cache_dict[this_uuid] = torch.zeros(1, 80, 0, 2)
+            # self.flow_cache_dict[this_uuid] = torch.zeros(1, 80, 0, 2)
         if source_speech_token.shape[1] == 0:
             p = threading.Thread(target=self.llm_job, args=(text, prompt_text, llm_prompt_speech_token, llm_embedding, this_uuid))
         else:
@@ -236,7 +236,7 @@ class CosyVoiceModel:
             self.llm_end_dict.pop(this_uuid)
             self.mel_overlap_dict.pop(this_uuid)
             self.hift_cache_dict.pop(this_uuid)
-            self.flow_cache_dict.pop(this_uuid)
+            # self.flow_cache_dict.pop(this_uuid)
         if torch.cuda.is_available():
             torch.cuda.empty_cache()
             torch.cuda.current_stream().synchronize()