Fine-tuning ruGPT-3 XL 1.3B to Modern LLM Level and Comparison with Gemma3 1B
An enthusiast fine-tuned the pretrain model ruGPT-3 XL 1.3B (from 2020) using SFT and DPO to modern LLM level. The resulting model is compared with Gemma3 1B from 2025; in many tests, ruGPT-3 XL showed more accurate and concise answers, especially in logic tasks and knowledge of Russian culture.
Сбербанк
Google/DeepMind
