Summary of Bggpt 1.0: Extending English-centric Llms to Other Languages, by Anton Alexandrov et al.
BgGPT 1.0: Extending English-centric LLMs to other languages
by Anton Alexandrov, Veselin Raychev, Dimitar I. Dimitrov, Ce Zhang, Martin Vechev, Kristina Toutanova
First submitted to arxiv on: 14 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents two new AI models, BgGPT-Gemma-2-27B-Instruct and BgGPT-Gemma-2-9B-Instruct, which are continually pretrained and fine-tuned versions of Google’s Gemma-2 models. These models are specifically designed for understanding and generating Bulgarian language content, leveraging over 100 billion tokens of Bulgarian and English text data. The authors demonstrate strong performance in Bulgarian language tasks, setting a new standard for language-specific AI models. To maintain the robust capabilities of the original Gemma-2 models, the authors incorporate continual learning strategies based on Branch-and-Merge techniques and thorough curation and selection of training data. The methodology is detailed, including the release of model weights with a commercial-friendly license, enabling broader adoption by researchers, companies, and hobbyists. Benchmarks are established using non-public educational data sources to evaluate models on Bulgarian language tasks as well as safety and chat capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper presents two new AI models that can understand and generate Bulgarian language content better than before. These models use a lot of text data from both Bulgarian and English languages. The authors show that these models perform very well in tasks like language translation, setting a new standard for AI models designed specifically for one language. To make sure the models don’t forget how to do other things, like understanding English, the authors used special techniques when training them. The paper also includes details on how to use the models and provides examples of how they can be tested. |
Keywords
» Artificial intelligence » Continual learning » Translation
