Open-source code for training and scaling up energy-based foundation models.
Data processing for and with foundation models! 🍎 🍋 🌽 ➡️ ➡️🍸 🍹 🍷
[NeurIPS 2026] Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models
Light-weight infrastructure for express LLM pre-training
Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities
Closed-loop fault localization and checkpoint recovery for LLM Pre-training
WB-WAM: Heterogeneous Body–Hand Pre-training for Humanoid Loco-Manipulation.