Update app.py

app.py

@@ -108,7 +108,7 @@ def  get_model(name_model):
     global models
     if name_model in   models:
         return models[name_model]
-    models[name_model]=VitsModel.from_pretrained(name_model,token=token).cuda()
+    models[name_model]=VitsModel.from_pretrained(name_model,token=token)
     models[name_model].decoder.apply_weight_norm()
     # torch.nn.utils.weight_norm(self.decoder.conv_pre)
     # torch.nn.utils.weight_norm(self.decoder.conv_post)
@@ -130,7 +130,7 @@ def   modelspeech(text=TXT,name_model="wasmdashai/vits-ar-sa-huba-v2",speaking_r
      model=get_model(name_model)
      model.speaking_rate=speaking_rate
      with torch.no_grad():
-        wav=list(_inference_forward_stream(model,input_ids=inputs.input_ids.cuda(),attention_mask=inputs.attention_mask.cuda(),speaker_embeddings= None,is_streaming=False))[0]
+        wav=list(_inference_forward_stream(model,input_ids=inputs.input_ids,attention_mask=inputs.attention_mask,speaker_embeddings= None,is_streaming=False))[0]
      # with torch.no_grad():
      #      wav = model(input_ids=inputs["input_ids"].cuda()).waveform.cpu().numpy().reshape(-1)#.detach()