add

app.py

@@ -106,19 +106,19 @@ class BaseTrainer(object):
        
         self.checkpoint_path = tmp_dir
         args.tmp_dir = tmp_dir
-        if self.rank == 0:
-            self.test_data = __import__(f"dataloaders.{args.dataset}", fromlist=["something"]).CustomDataset(args, "test")
-            self.test_loader = torch.utils.data.DataLoader(
-                self.test_data, 
-                batch_size=1,  
-                shuffle=False,  
-                num_workers=args.loader_workers,
-                drop_last=False,
-            )
+
+        self.test_data = __import__(f"dataloaders.{args.dataset}", fromlist=["something"]).CustomDataset(args, "test")
+        self.test_loader = torch.utils.data.DataLoader(
+            self.test_data, 
+            batch_size=1,  
+            shuffle=False,  
+            num_workers=args.loader_workers,
+            drop_last=False,
+        )
         logger.info(f"Init test dataloader success")
-        model_module = __import__(f"models.{args.model}", fromlist=["something"])
-        
-        self.model = getattr(model_module, args.g_name)(args).cuda()
+
+        from models.denoiser import MDM
+        self.model = MDM(args).cuda()
         
         if self.rank == 0:
             logger.info(self.model)
@@ -768,7 +768,8 @@ demo = gr.Interface(
     ],
     title='SynTalker: Enabling Synergistic Full-Body Control in Prompt-Based Co-Speech Motion Generation',
     description="1. Upload your audio.  <br/>\
-        2. Then, sit back and wait for the rendering to happen! This may take a while (e.g. 2 minutes) <br/>\
+        2. Then, sit back and wait for the rendering to happen! This may take a while (e.g. 2-5 minutes) <br/>\
+        (The reason of running time so long is that provided GPU have an limitation in GPU running time, we must use CPU to handle some GPU tasks)\
         3. After, you can view the videos.  <br/>\
         4. Notice that we use a fix face animation, our method only produce body motion. <br/>\
         5. Use DDPM sample strategy will generate a better result, while it will take more inference time.  \