This is an more than 1000X faster LSTM-CRF implementation modified from the slower version in the Pytorch official tutorial (URL: https://pytorch.org/tutorials/beginner/nlp/advanced_tutorial.html ).
I have modified the dynamic planning part, including Viterbi decoding and partition function calculation. In the experiment, it has achieved a speed increase of more than 50 times compared to the original version. Furthermore, the original version can only input one sample at a time when it is trained. In the most recent updated module 'lstm-crf-parallel.py', I modified the model to support parallel computing for batch, so that the training time was greatly improved again. When the batchsize is large, parallel computing can bring you hundreds of times faster.
You only need to call the function 'neg_log_likelihood_parallel' in module 'lstm-crf-parallel.py', instead of the original 'neg_log_likelihood', to perform parallel training when the loss is calculated.
It is important to note that in the parallel module 'lstm-crf-parallel.py', the input tensor is shaped as [batchSize,timeStep,embedding], while in the original version, the input tensor is shaped as [timeStep,batchSize,embedding] github地址： https://github.com/mali19064/LSTM-CRF-pytorch-faster