Pre-training Research Engineer
Jobverse.io · San Francisco, CA · Today
EngineeringFull-time
Sciforium is an AI infrastructure company developing next-generation multimodal AI models and a proprietary high-efficiency serving platform. The Pre-training Research Engineer will implement, scale, and improve byte-native and multimodal foundation models while developing production-grade training code and infrastructure. The role focuses on model architecture, training objectives, optimization, scaling experiments, and analysis of model behavior and quality. Responsibilities Train large byte-native and multimodal foundation models across massive, heterogeneous corporaImplement and evaluate new model architectures, training objectives, and optimization methodsDevelop stable pre-training recipes and run scaling experiments for novel architecturesConduct ablations and analyze training dynamics, model behavior, and base-model qualityWork with data and distributed training engineers to improve training efficiency, reliability, and scalability