I am learning quantization of LSTM in Pytorch. from the source code I can not see the implementation of “static” quantization of LSTM, the last function I can see is as follows:torch.quantized_lstm, I can not see the implementation of this function.By the way, are there any paper about dynamic quantization of LSTM ?
Thanks.
I’m not aware of papers about dynamic quantization for LSTM. here is an example for statically quantize LSTM: https://github.com/pytorch/pytorch/blob/main/test/quantization/fx/test_quantize_fx.py#L4782 in fx graph mode quantization, dynamically quantize LSTM examples can be found here: https://github.com/pytorch/pytorch/blob/main/test/quantization/eager/test_quantize_eager_ptq.py#L1333 (with eager mode quantization)