Elamur Research⌕
☰
home / papers / 2026 / From Standard Transformers to Modern LLMs: Bringing Dialogue Models, RAG, and Agents to the Classroom

From Standard Transformers to Modern LLMs: Bringing Dialogue Models, RAG, and Agents to the Classroom

Paper · 2026 · Maria Tikhonova, Viktoriia A. Chekalina, Artem Chervyakov et al.

Summary

This TeachNLP 2026 paper describes the second edition of an open university course on Transformer-based LLMs and multimodal models, first presented by Nikishina et al. (2024). The authors argue that knowing Transformer internals no longer prepares students for industry work. Practitioners now mostly assemble complete LLM systems: they ground answers through retrieval, coordinate tools with agents, call models through APIs, and manage serving costs. The revised curriculum has 14 sessions for learners who know Python and ML/DL but have little experience with LLMs. Each session pairs a lecture with Jupyter notebook exercises. New or reworked material covers:

  • modern architectural components (RoPE, GQA/MQA);
  • long-context modeling and KV-cache efficiency;
  • RAG pipelines;
  • ReAct-style agents;
  • an API-based practical track;
  • uncertainty estimation for white-box and black-box LLMs;
  • multimodal dialogue grounding with vLLM-based inference.

The paper reports no empirical evaluation. Its contribution is a modular, openly released set of teaching materials that instructors can adopt in full or in part.

Key points

  • The course has 14 sessions and is hosted on GitHub (s-nlp/transformers-course). Slides and notebooks are modular, and homework and quizzes are available from the authors on request.
  • Session 7 (RAG) compares frozen RAG, RePlug-style augmentation and classical RAG (Lewis et al., 2020). Session 8 has students build ReAct agents with tool wrappers and stopping criteria, and analyze failures such as looping and brittle parsing.
  • The API track in Sessions 7–8 uses GigaChat and covers authentication, rate limits, error handling and cost-aware usage.
  • Session 6 covers long-context problems such as KV-cache memory and needle-in-a-haystack retrieval, along with Transformer-XL, RMT and a Russian long-context benchmark. Sessions 9–10 cover uncertainty estimation, with practical work in LM-Polygraph.
  • Session 12 covers multimodal dialogue grounding, with vLLM for efficient serving and MERA Multi for evaluation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓National Research University Higher School of EconomicsMoscow · Russia23 works · 39,585 citations🎓Lomonosov Moscow State UniversityMoscow · Russia14 works · 2,430 citations🤝AIRI - Artificial Intelligence Research InstituteMoscow · Russia5 works · 29 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Dialogue systems · 624 worksVision & language · 657 works

Cite

Cite as Maria Tikhonova, Viktoriia A. Chekalina, Artem Chervyakov, Alexey Zaytsev and Alexander Panchenko. 2026. From Standard Transformers to Modern LLMs: Bringing Dialogue Models, RAG, and Agents to the Classroom. Proceedings of the Seventh Workshop on Teaching Natural Language Processing (TeachNLP 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.teachingnlp-1.8

@inproceedings{tikhonova2026standard,
  title = {{From Standard Transformers to Modern LLMs: Bringing Dialogue Models, RAG, and Agents to the Classroom}},
  author = {Maria Tikhonova, Viktoriia A. Chekalina, Artem Chervyakov, Alexey Zaytsev and Alexander Panchenko},
  booktitle = {Proceedings of the Seventh Workshop on Teaching Natural Language Processing (TeachNLP 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.teachingnlp-1.8},
  url = {https://doi.org/10.18653/v1/2026.teachingnlp-1.8}
}

References

5 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aydar Bulatov, Yury Kuratov, and Mikhail Burtsev. 2022. Recurrent memory transformer. In Advances in Neural Information Processing Systems, volume 35. find
  2. Irina Nikishina, Maria Tikhonova, Viktoriia Chekalina, Alexey Zaytsev, Artem Vazhentsev, and Alexander Panchenko. 2024. Industry vs academia: Running a course on transformers in two setups. In Proceedings of the Sixth Workshop on Teaching NLP. Association for Computational Linguistics. find
  3. Artem Chervyakov, Ulyana Isaeva, Anton Emelyanov, Artem Safin, Maria Tikhonova, Alexander Kharitonov, Yulia Lyakh, Petr Surovtsev, Denis Shevelev, Vildan Saburov, Vasily Konovalov, Elisei Rykov, Ivan Sviridov, Amina Miftakhova, Ilseyar Weijia Shi, Sewon Min, Michihiro Yasunaga, Minjoon Seo, Richard… — REPLUG: Retrieval-Augmented Black-Box Language Models
  4. Roman Vashurin, Ekaterina Fadeeva, Artem Vazhentsev, Lyudmila Rvanova, Daniil Vasilev, Akim Tsvigun, Sergey Petrakov, Rui Xing, Abdelrahman Sadallah, Kirill Grishchenkov, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, Maxim Panov, and Artem Shelmanov. 2025. Benchmarking uncertainty quantificat…
  5. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. ReAct: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tikhonova-2026-standard-transformers-modern-llms-bringing/