@Pouakai There is a Claude skill to improve LLM training scripts: https://github.com/allenai/OLMo-core/blob/main/.claude/skills/training-smoke-test/SKILL.md They use different training data but seem to generate example configuration for their training process. This is like opening a new restaurant by hiring McDonalds workers to "reverse-engineer" ingredients for Vkusno i Tochka IMO