FastWAM (kimtaey/fastwam-lpl-rt-a4)

FastWAM is a world action model for robot control that uses video-based world modeling during training but performs direct action prediction at inference time, bypassing future-frame generation. This checkpoint (lpl-rt-a4) is a fine-tuned variant published by Hugging Face user kimtaey, part of the broader FastWAM model family. Source: supabase.