Update app.py

app.py

@@ -40,6 +40,86 @@ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 #                         discriminator=False,
 #                         duration=False
 # )
+class model_onxx:
+    def __init__(self):
+        self.model=None
+        self.n_onxx=""
+        pass
+    def function_change(self,n_model,token,n_onxx,choice):
+        if choice=="decoder":
+
+            V=self.convert_model_decoder_onxx(n_model,token,n_onxx)
+        elif choice=="all only decoder":
+            V=self.convert_model_decoder_onxx(n_model,token,n_onxx)
+        else:
+            V=self.convert_to_onnx_all(n_model,token,n_onxx)
+        return V
+
+    def install_model(self,n_model,token,n_onxx):
+        self.n_onxx=n_onxx
+        self.model= VitsModel.from_pretrained(n_model,token=token)
+        return self.model
+    def convert_model_decoder_onxx(self,n_model,token,namemodelonxx):
+        self.model= VitsModel.from_pretrained(n_model,token=token)
+        x=f"{namemodelonxx}.onnx"
+        return x
+    def convert_to_onnx_only_decoder(self,n_model,token,namemodelonxx):
+          model=VitsModel.from_pretrained(n_model,token=token)
+          x=f"{namemodelonxx}.onnx"
+          vocab_size = model.text_encoder.embed_tokens.weight.size(0)
+          example_input = torch.randint(0, vocab_size, (1, 100), dtype=torch.long)
+          torch.onnx.export(
+              model,  # The model to be exported
+              example_input,  # Example input for the model
+              x,  # The filename for the exported ONNX model
+              opset_version=11,  # Use an appropriate ONNX opset version
+              input_names=['input'],  # Name of the input layer
+              output_names=['output'],  # Name of the output layer
+              dynamic_axes={
+                  'input': {0: 'batch_size', 1: 'sequence_length'},  # Dynamic axes for variable-length inputs
+                  'output': {0: 'batch_size'}
+              }
+           )
+          return x
+    def convert_to_onnx_all(self,n_model,token ,namemodelonxx):
+
+          model=VitsModel.from_pretrained(n_model,token=token)
+          x=f"{namemodelonxx}.onnx"
+
+          vocab_size = model.text_encoder.embed_tokens.weight.size(0)
+          example_input = torch.randint(0, vocab_size, (1, 100), dtype=torch.long)
+          torch.onnx.export(
+              model,  # The model to be exported
+              example_input,  # Example input for the model
+              x,  # The filename for the exported ONNX model
+              opset_version=11,  # Use an appropriate ONNX opset version
+              input_names=['input'],  # Name of the input layer
+              output_names=['output'],  # Name of the output layer
+              dynamic_axes={
+                  'input': {0: 'batch_size', 1: 'sequence_length'},  # Dynamic axes for variable-length inputs
+                  'output': {0: 'batch_size'}
+              }
+           )
+          return x
+    def starrt(self):
+        #with gr.Blocks() as demo:
+            with gr.Row():
+                  with gr.Column():
+                    text_n_model=gr.Textbox(label="name model")
+                    text_n_token=gr.Textbox(label="token")
+                    text_n_onxx=gr.Textbox(label="name model onxx")
+                    choice = gr.Dropdown(choices=["decoder", "all anoly decoder", "All"], label="My Dropdown")
+
+                  with gr.Column():
+
+                    btn=gr.Button("convert")
+                    label=gr.Label("return  name model onxx")
+                    btn.click(self.function_change,[text_n_model,text_n_token,text_n_onxx,choice],[label])
+                    #choice.change(fn=function_change, inputs=choice, outputs=label)
+        #return demo
+c=model_onxx()
+#cc=c.starrt()
+###############################################################
 Lst=['input_ids',
  'attention_mask',
  'waveform',
@@ -690,6 +770,8 @@ with gr.Blocks() as interface:
         output_ini = gr.Textbox(label="token")
         label=gr.Label("hhh")
         btn_init.click(loadd_d,[output_i],[label])
+    with gr.Accordion("read model ", open=False):
+         c.starrt()
     with gr.Accordion("init_Starting ", open=False):
         btn_init = gr.Button("init start")
         output_init = gr.Textbox(label="init")