ParScale

Base models trained on 1T high-quality tokens, demonstrating strong competitiveness among existing SOTA small models (<2B).

Instruct models from the ParScale-1.8B base models, trained on SmolTalk-1M to enable conversational capabilities.