Merge pull request #603 from oongrider/f_free_cache_dict

release flow_cache_dict to prevent potential memory leaks in long-running processes

Author: aluminumbox

cosyvoice/cli/model.py

@@ -201,6 +201,7 @@ def tts(self, text, flow_embedding, llm_embedding=torch.zeros(0, 192),
             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)
 
     def vc(self, source_speech_token, flow_prompt_speech_token, prompt_speech_feat, flow_embedding, stream=False, speed=1.0, **kwargs):
         # this_uuid is used to track variables related to this inference thread