language:
- ru
distilrubert-tiny-cased-conversational
Conversational DistilRuBERT-tiny (Russian, cased, 2‑layer, 768‑hidden, 12‑heads, 107M parameters) was trained on OpenSubtitles[1], Dirty, Pikabu, and a Social Media segment of Taiga corpus[2] (as Conversational RuBERT). It can be considered as tiny copy of our Conversational DistilRuBERT-base
Our DistilRuBERT-tiny was highly inspired by [3], [4]. Namely, we used
- KL loss (between teacher and student output logits)
- MLM loss (between tokens labels and student output logits)
- Cosine embedding loss (between mean of six consecutive hidden states from teacher's encoder and one hidden state of the student)
- MSE loss (between six consecutive attention maps from teacher's encoder and one attention map of the student)
[1]: P. Lison and J. Tiedemann, 2016, OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles. In Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC 2016)
[2]: Shavrina T., Shapovalova O. (2017) TO THE METHODOLOGY OF CORPUS CONSTRUCTION FOR MACHINE LEARNING: «TAIGA» SYNTAX TREE CORPUS AND PARSER. in proc. of “CORPORA2017”, international conference , Saint-Petersbourg, 2017.
[3]: Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019). DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108.
[4]: https://github.com/huggingface/transformers/tree/master/examples/research_projects/distillation