Trying ONNX models on CPU.

app.py

@@ -28,8 +28,8 @@ print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 # tokenizer_name = "MoritzLaurer/deberta-v3-base-zeroshot-v2.0"
 model_name = "MoritzLaurer/ModernBERT-large-zeroshot-v2.0"
 
-classifier_cpu = pipeline(task="zero-shot-classification", model=model_name)
-classifier_gpu = pipeline(task="zero-shot-classification", model=model_name, device="cuda:0")
+classifier_cpu = pipeline(task="zero-shot-classification", model=model_name, provider="CPUExecutionProvider")
+classifier_gpu = pipeline(task="zero-shot-classification", model=model_name, provider="CUDAExecutionProvider", device="cuda:0")
 
 def classify(data_string, request: gradio.Request):
     if request: