Merge pull request #1187 from hexisyztem/dev/Comet

Refactor CUDA stream context management in CosyVoice2Model

Author: aluminumbox

cosyvoice/cli/model.py

@@ -385,7 +385,7 @@ def tts(self, text, flow_embedding, llm_embedding=torch.zeros(0, 192),
         # this_uuid is used to track variables related to this inference thread
         self.synchronize_stream()
         stream_context = self.stream_context_pool.get()
-        with torch.cuda.stream(stream_context):
+        with stream_context:
 
             this_uuid = str(uuid.uuid1())
             with self.lock: