• FAQ Skill for Russian & English with a tutorial
  • Improved Seq2Seq GO Bot: attention over knowledge base, encoder attention, embeddings instead of one-hots
  • Pretrainded model for Seq2Seq GO Bot on driver dataset with BLEU=13.2
  • Improved ODQA model with squad logits: f1 raised from 14 to 26
  • Hashing tf-idf vectorizer component
  • Text classifiers now compatible with ELMO embedder
  • Improved documentation
  • Bug fixes and minor improvements