WenyuanLi212

H800/L20 显卡推理首包时延约1s，有什么优化手段吗？

2

front.app代码加了测时延的部分： def frontend_zero_shot(self, tts_text, prompt_text, prompt_wav, resample_rate, zero_shot_spk_id): tts_text_token, tts_text_token_len = self._extract_text_token(tts_text) if zero_shot_spk_id == '': time_1 = time.time() prompt_text_token, prompt_text_token_len = self._extract_text_token(prompt_text) time_2 = time.time() speech_feat, speech_feat_len = self._extract_speech_feat(prompt_wav)...