Trying ONNX models on CPU.

app.py

@@ -21,15 +21,14 @@ app.add_middleware(
 )
 
 print(f"Is CUDA available: {torch.cuda.is_available()}")
-print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
-
-# "xenova/mobilebert-uncased-mnli" "typeform/mobilebert-uncased-mnli" Fast but small--same as bundled in Statosphere
+if torch.cuda.is_available():
+    print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 
 model_name = "MoritzLaurer/roberta-large-zeroshot-v2.0-c"
 tokenizer_name = "MoritzLaurer/roberta-large-zeroshot-v2.0-c"
 
 classifier_cpu = pipeline(task="zero-shot-classification", model=model_name, tokenizer=tokenizer_name)
-classifier_gpu = pipeline(task="zero-shot-classification", model=model_name, tokenizer=tokenizer_name, device="cuda:0")
+classifier_gpu = pipeline(task="zero-shot-classification", model=model_name, tokenizer=tokenizer_name, device="cuda:0") if torch.cuda.is_available() else classifier_cpu
 
 def classify(data_string, request: gradio.Request):
     if request: