Update app.py

app.py

@@ -565,29 +565,8 @@ dir_model='wasmdashai/vits-ar-huba-fine'
 
 global_step=0
 wandb.login(key= "782b6a6e82bbb5a5348de0d3c7d40d1e76351e79")
-parser = HfArgumentParser((ModelArguments, DataTrainingArguments, VITSTrainingArguments))
-json_file = os.path.abspath('VitsModelSplit/finetune_config_ara.json')
-model_args, data_args, training_args = parser.parse_json_file(json_file = json_file)
-print('start')
-sgl=get_state_grad_loss(mel=True,
-                       # generator=False,
-                    #    discriminator=False,
-                        duration=False)
-
-print(training_args)
-training_args.num_train_epochs=1000
-training_args.fp16=True
-training_args.eval_steps=300
-training_args.weight_kl=1
-training_args.d_learning_rate=2e-4
-training_args.learning_rate=2e-4
-training_args.weight_mel=45
-training_args.num_train_epochs=4
-training_args.eval_steps=1000
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-print(device)
 
-b=int(id)
+
 ctrain_datasets,eval_dataset,full_generation_dataset=get_data_loader(train_dataset_dirs = train_dataset_dirs,
                                                           eval_dataset_dir = os.path.join(dataset_dir,'eval'),
                                                         full_generation_dir = os.path.join(dataset_dir,'full_generation'),
@@ -602,13 +581,36 @@ print('loadeed')
 @spaces.GPU
 def greet(text,id):
     global GK 
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, VITSTrainingArguments))
+    json_file = os.path.abspath('VitsModelSplit/finetune_config_ara.json')
+    model_args, data_args, training_args = parser.parse_json_file(json_file = json_file)
+    print('start')
+    sgl=get_state_grad_loss(mel=True,
+                           # generator=False,
+                        #    discriminator=False,
+                            duration=False)
+    
+    print(training_args)
+    training_args.num_train_epochs=1000
+    training_args.fp16=True
+    training_args.eval_steps=300
+    training_args.weight_kl=1
+    training_args.d_learning_rate=2e-4
+    training_args.learning_rate=2e-4
+    training_args.weight_mel=45
+    training_args.num_train_epochs=4
+    training_args.eval_steps=1000
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    print(device)
+
 
     for i in range(10000):
 # model.train(True)
         print(f'clcye epochs ={i}')
         yield f'clcye epochs ={i}'
         
-        # model.setMfA(monotonic_align.maximum_path)
+        model=VitsModel.from_pretrained(dir_model,token=token).to("cuda")
+
         #dir_model_save=dir_model+'/vend'