« All posts

LoongForge: High-Performance Training System for Embodied Models

LoongForge-Embodied offers a PyTorch-based training subsystem for VLA and WAM models.

LoongForge-Embodied introduces a dedicated training subsystem for Vision-Language-Action (VLA) and world-action models. Built on a PyTorch DDP/FSDP engine, this system focuses on data plumbing and iteration speed rather than model parameter scale.

This synthesis was produced from its source by AI; there is no human editor or manual review step. How we work