【fast_inference_】通过简单添加torch.no_grad()修饰器，可能会使速度加快 (#930)

* 添加with torch.no_grad()，速度快一大截

* 恢复先前缩进

* 恢复make batch的位置

* 改用修饰器

* 去除没必要的增加的空行

---------

Co-authored-by: XTer <xxoy1234@outlook.com>

.gitignore

@@ -10,6 +10,8 @@ reference
 GPT_weights
 SoVITS_weights
 TEMP
+PortableGit
 ffmpeg.exe
 ffprobe.exe
-
+tmp_audio
+trained

GPT_SoVITS/TTS_infer_pack/TTS.py

@@ -249,8 +249,6 @@ class TTS:
         if self.configs.is_half and str(self.configs.device)!="cpu":
             self.bert_model = self.bert_model.half()
         
-        
-
     def init_vits_weights(self, weights_path: str):
         print(f"Loading VITS weights from {weights_path}")
         self.configs.vits_weights_path = weights_path
@@ -437,7 +435,8 @@ class TTS:
                  device:torch.device=torch.device("cpu"),
                  precision:torch.dtype=torch.float32,
                  ):
-        
+        # 但是这里不能套，反而会负优化
+        # with torch.no_grad():
         _data:list = []
         index_and_len_list = []
         for idx, item in enumerate(data):
@@ -485,6 +484,8 @@ class TTS:
             norm_text_batch = []
             bert_max_len = 0
             phones_max_len = 0
+            # 但是这里也不能套，反而会负优化
+            # with torch.no_grad():
             for item in item_list:
                 if prompt_data is not None:
                     all_bert_features = torch.cat([prompt_data["bert_features"], item["bert_features"]], 1)\
@@ -568,7 +569,8 @@ class TTS:
         '''
         self.stop_flag = True
     
-    
+    # 使用装饰器
+    @torch.no_grad()
     def run(self, inputs:dict):
         """
         Text to speech inference.
@@ -643,11 +645,10 @@ class TTS:
             ((self.prompt_cache["prompt_semantic"] is None) or (self.prompt_cache["refer_spec"] is None)):
             raise ValueError("ref_audio_path cannot be empty, when the reference audio is not set using set_ref_audio()")
 
-
         ###### setting reference audio and prompt text preprocessing ########
         t0 = ttime()
         if (ref_audio_path is not None) and (ref_audio_path != self.prompt_cache["ref_audio_path"]):
-                self.set_ref_audio(ref_audio_path)
+            self.set_ref_audio(ref_audio_path)
 
         if not no_prompt_text:
             prompt_text = prompt_text.strip("\n")
@@ -664,7 +665,6 @@ class TTS:
                 self.prompt_cache["bert_features"] = bert_features
                 self.prompt_cache["norm_text"] = norm_text
 
-        
         ###### text preprocessing ########
         t1 = ttime()
         data:list = None
@@ -718,6 +718,7 @@ class TTS:
                             )
                 return batch[0]
 
+
         t2 = ttime()
         try:
             print("############ 推理 ############")
@@ -745,18 +746,18 @@ class TTS:
                 else:
                     prompt = self.prompt_cache["prompt_semantic"].expand(len(all_phoneme_ids), -1).to(self.configs.device)
 
-                with torch.no_grad():
-                    pred_semantic_list, idx_list = self.t2s_model.model.infer_panel(
-                        all_phoneme_ids,
-                        all_phoneme_lens,
-                        prompt,
-                        all_bert_features,
-                        # prompt_phone_len=ph_offset,
-                        top_k=top_k,
-                        top_p=top_p,
-                        temperature=temperature,
-                        early_stop_num=self.configs.hz * self.configs.max_sec,
-                    )
+
+                pred_semantic_list, idx_list = self.t2s_model.model.infer_panel(
+                    all_phoneme_ids,
+                    all_phoneme_lens,
+                    prompt,
+                    all_bert_features,
+                    # prompt_phone_len=ph_offset,
+                    top_k=top_k,
+                    top_p=top_p,
+                    temperature=temperature,
+                    early_stop_num=self.configs.hz * self.configs.max_sec,
+                )
                 t4 = ttime()
                 t_34 += t4 - t3
 
@@ -765,6 +766,9 @@ class TTS:
 
                 batch_audio_fragment = []
 
+                # 这里要记得加 torch.no_grad() 不然速度慢一大截
+                # with torch.no_grad():
+            
                 # ## vits并行推理 method 1
                 # pred_semantic_list = [item[-idx:] for item, idx in zip(pred_semantic_list, idx_list)]
                 # pred_semantic_len = torch.LongTensor([item.shape[0] for item in pred_semantic_list]).to(self.configs.device)
@@ -791,7 +795,6 @@ class TTS:
                 audio_frag_end_idx.insert(0, 0)
                 batch_audio_fragment= [_batch_audio_fragment[audio_frag_end_idx[i-1]:audio_frag_end_idx[i]] for i in range(1, len(audio_frag_end_idx))]
 
-
                 # ## vits串行推理
                 # for i, idx in enumerate(idx_list):
                 #     phones = batch_phones[i].unsqueeze(0).to(self.configs.device)