Efficient Sampled Softmax for Tensorflow
Abstract
This short paper discusses an efficient implementation of sampled softmax loss for Tensorflow. The speedup over the default implementation is achieved due to simplification of the graph for the forward and backward passes.
0