我如何从 python 中的预训练 model 中获取重量并在 tensorflow 中使用它？

Question

How can I get weights from a pre-trained model in PyTorch and use it in TensorFlow?

this is the pre-trained model:

lstm = torch.hub.load("BruceWen120/medal", "lstm")

Answer 1

It is as of now not possible to convert PyTorch code into Tensorflow. (That is a transpiler to convert code written to train in one framework into another is not available). The reason is because training code is written in different ways in both libraries.

However if model trained in one library is available, you can use it in the other. The reason in that neural networks use standardized components which can be 1-to-1 corresponded between different frameworks.

ONNX: Open Neural Network Exchange Format is a bridge format to transfer trained models between libraries.

While PyTorch supports onnx out of the box, Tensorflow can be connected by a open source connector too. ONNX: Open Neural Network Exchange Format