ULMFiT - Russian

Hi!

I have published a language model trained on a newspaper subset of Taiga corpus. You can get it here:

As mentioned previously, it achieves 21.98 perplexity on a 20 million token validation set.

1 Like