eval-fastwam

Deploy and evaluate FastWAM (Fast World Action Model) on the LIBERO and LIBERO-Plus benchmarks. Covers environment setup, checkpoint download, Wan2.2 model preparation, single-GPU and multi-GPU evaluation, per-task and per-category result aggregation. Use this skill whenever someone needs to install FastWAM, run inference on LIBERO tasks, debug model loading issues (CUDA version mismatch, HF repo redirects, ActionDiT preprocessing), parallelize evaluation across GPUs, or compare FastWAM results against other VLA models like LingBot-VA.