search

yandex / faster-rnnlm

Faster Recurrent Neural Network Language Modeling Toolkit with Noise Contrastive Estimation and Hierarchical Softmax
Other
562 stars 138 forks source link

One Billion Word Benchmark precision #40

Open alexnanchen opened 7 years ago

alexnanchen commented 7 years ago

Hello,

First thanks for this nice benchmark!

I have a question regarding the last figure. For the "gru-insyn+direct+KN" results, how do you combine KN with gru-insyn+direct? Is it during training? Do you combine log probabilities of the two models during evaluation with some kind of interpolation?

Many thanks!

Alexandre Nanchen