daVinci-LLM bridges the gap between industry resources and academic research to systematically study large language model pretraining with full transparency.

daVinci-LLM combines industrial-scale computational resources with complete research freedom to address the under-explored science of LLM pretraining

The project adopts a fully-open paradigm, releasing complete data processing pipelines, full training processes, and systematic exploration results

Uses the Data Darwinism framework with principled L0-L9 taxonomy to establish systematic methodology for data processing, which the field currently lacks

Addresses a critical structural paradox where well-resourced companies lack research transparency while academic institutions have freedom but lack pretraining-scale compute