Differences
Hi, I want to know the difference between onnx and onnx_data file.
I also wonder the difference between decoder_model.onnx and decoder_with_past_model.onnx.
Thanks
Hi, I want to know the difference between onnx and onnx_data file.
I also wonder the difference between decoder_model.onnx and decoder_with_past_model.onnx.
Thanks
The .onnx file contains the model's computation graph, which can be understood as the model's "outline" or "framework". The .onnx_data contains the model's weights.
When generating text, Transformer models utilize calculations from previous steps, specifically the keys and values from self-attention. Models with "with past" will use these "historical information" for subsequent computations, whereas models without "with past" won't.
The difference between them will be reflected in their input nodes.