Conversational DistilRuBERT-tiny (Russian, cased, 3‑layers, 264‑hidden, 12‑heads, 10.4M 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 Conversational DistilRuBERT-small.

Our DistilRuBERT-tiny is highly inspired by [3], [4] and architecture is very close to [5]. Namely, we use

The key features are:

Here is comparison between teacher model (Conversational RuBERT) and other distilled models.

DistilRuBERT-tiny was trained for about 100 hrs. on 7 nVIDIA Tesla P100-SXM2.0 16Gb.

We used PyTorchBenchmark from transformers to evaluate model's performance and compare it with other pre-trained language models for Russian. All tests were performed on Intel(R) Xeon(R) CPU E5-2698 v4 @ 2.20GHz and nVIDIA Tesla P100-SXM2.0 16Gb.

| Model name | Batch size | Seq len | Time, s || Mem, MB || |---|---|---|------||------|| | | | | CPU | GPU | CPU | GPU | | rubert-base-cased-conversational | 1 | 512 | 0.147 | 0.014 | 897 | 1531 | | distilrubert-base-cased-conversational | 1 | 512 | 0.083 | 0.006 | 766 | 1423 | | distilrubert-small-cased-conversational | 1 | 512 | 0.03 | 0.002 | 600 | 1243 | | cointegrated/rubert-tiny | 1 | 512 | 0.041 | 0.003 | 272 | 919 | | distilrubert-tiny-cased-conversational | 1 | 512 | 0.023 | 0.003 | 206 | 855 | | rubert-base-cased-conversational | 16 | 512 | 2.839 | 0.182 | 1499 | 2071 | | distilrubert-base-cased-conversational | 16 | 512 | 1.065 | 0.055 | 2541 | 2927 | | distilrubert-small-cased-conversational | 16 | 512 | 0.373 | 0.003 | 1360 | 1943 | | cointegrated/rubert-tiny | 16 | 512 | 0.628 | 0.004 | 1293 | 2221 | | distilrubert-tiny-cased-conversational | 16 | 512 | 0.219 | 0.003 | 633 | 1291 |

To evaluate model quality, we fine-tuned DistilRuBERT-tiny on classification (RuSentiment, ParaPhraser), NER and question answering data sets for Russian and obtained scores very similar to the Conversational DistilRuBERT-small.

If you found the model useful for your research, we are kindly ask to cite this paper:

@misc{https://doi.org/10.48550/arxiv.2205.02340,
  doi = {10.48550/ARXIV.2205.02340},
  
  url = {https://arxiv.org/abs/2205.02340},
  
  author = {Kolesnikova, Alina and Kuratov, Yuri and Konovalov, Vasily and Burtsev, Mikhail},
  
  keywords = {Computation and Language (cs.CL), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
  
  title = {Knowledge Distillation of Russian Language Models with Reduction of Vocabulary},
  
  publisher = {arXiv},
  
  year = {2022},
  
  copyright = {arXiv.org perpetual, non-exclusive license}
}

[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

[5]: https://habr.com/ru/post/562064/, https://huggingface.co/cointegrated/rubert-tiny