app.py

import gradio as gr
import spaces
import torch
from transformers import AutoTokenizer,VitsModel
import os
  

token=os.environ.get("key_")
tokenizer = AutoTokenizer.from_pretrained("wasmdashai/vtk",token=token)
models= {}
@spaces.GPU
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].decoder.apply_weight_norm()
    # torch.nn.utils.weight_norm(self.decoder.conv_pre)
    # torch.nn.utils.weight_norm(self.decoder.conv_post)
    for flow in models[name_model].flow.flows:
        torch.nn.utils.weight_norm(flow.conv_pre)
        torch.nn.utils.weight_norm(flow.conv_post)
    return models[name_model]


zero = torch.Tensor([0]).cuda()
print(zero.device) # <-- 'cpu' 🤔
import torch
@spaces.GPU
def   modelspeech(text,name_model):
     
    
     inputs = tokenizer(text, return_tensors="pt")
     model=get_model(name_model)
     with torch.no_grad():
          wav = model(input_ids=inputs["input_ids"].cuda()).waveform.cpu().numpy().reshape(-1)#.detach()
          
     return  model.config.sampling_rate,wav#remove_noise_nr(wav)

model_choices = gr.Dropdown(
                            choices=[
                                "wasmdashai/vits-ar",
                                "wasmdashai/vits-ar-sa-huba",
                                "wasmdashai/vits-ar-sa-ms",
                                "wasmdashai/vits-ar-sa-magd",
                                "wasmdashai/vtk",
                            ],
                            label="اختر النموذج",
                            value="wasmdashai/vtk",
                        )
demo = gr.Interface(fn=modelspeech, inputs=["text",model_choices], outputs=["audio"])
demo.queue()
demo.launch()