ONNX推理warning: TensorRT does not natively support INT64. Attempting to cast down to INT32.

只想用ONNX进行模型推理,加载时报Warning,加载模型时间也特别长。

加载模型的代码:

self.session = onnxruntime.InferenceSession(model_path, providers=onnxruntime.get_available_providers())

修改为:

self.session = onnxruntime.InferenceSession(model_path, providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

你可能感兴趣的:(人工智能)