LoongForge: Baidu Open-Sources High-Performance Training Framework for LLM, VLM, VLA, and Video Models

GitHub / Hacker News·low signal

Baidu's Baige team released LoongForge, an open-source training framework supporting LLMs, vision-language models (VLM), vision-language-action models (VLA), and video generation models like Wan. The Show HN post hit 10 points. For builders exploring multimodal training beyond text, this is a significant Chinese AI infrastructure contribution — particularly the VLA support, which is relevant to robotics and embodied AI pipelines.