add vllm example

cosyvoice/cli/cosyvoice.py

@@ -224,8 +224,6 @@ class CosyVoice3(CosyVoice2):
                         '{}/hift.pt'.format(model_dir))
         if load_vllm:
             self.model.load_vllm('{}/vllm'.format(model_dir))
-        if load_jit:
-            self.model.load_jit('{}/flow.encoder.{}.zip'.format(model_dir, 'fp16' if self.fp16 is True else 'fp32'))
         if load_trt:
             self.model.load_trt('{}/flow.decoder.estimator.{}.mygpu.plan'.format(model_dir, 'fp16' if self.fp16 is True else 'fp32'),
                                 '{}/flow.decoder.estimator.fp32.onnx'.format(model_dir),

vllm_example.py

@@ -21,8 +21,7 @@ def cosyvoice2_example():
 def cosyvoice3_example():
     """ CosyVoice3 vllm usage
     """
-    cosyvoice = CosyVoice3('pretrained_models/CosyVoice3-0.5B', load_jit=True, load_trt=True, load_vllm=True, fp16=True)
-    prompt_speech_16k = load_wav('./asset/zero_shot_prompt.wav', 16000)
+    cosyvoice = CosyVoice3('pretrained_models/CosyVoice3-0.5B', load_trt=True, load_vllm=True, fp16=True)
     for i in tqdm(range(100)):
         set_all_random_seed(i)
         for _, _ in enumerate(cosyvoice.inference_zero_shot('收到好友从远方寄来的生日礼物，那份意外的惊喜与深深的祝福让我心中充满了甜蜜的快乐，笑容如花儿般绽放。', '希望你以后能够做的比我还好呦。', './asset/zero_shot_prompt.wav', stream=False)):