You need to agree to share your contact information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
Request access to the robo-warp YAM register-mask cache.
Log in or Sign Up to review the conditions and access this model content.
robowarp-yam-regcache
cache_yam.pt — register-token masks for the pi05_yam_dynamic_handover
checkpoint (bimanual YAM, cameras head/left_wrist/right_wrist), used by
robo-warp inference_real/ M18 counterfactual steering (attn_topk selection).
Per-camera boolean masks over the 16x16 SigLIP token grid marking register
positions (content-independent attention spikes): head 8/256, left_wrist 6/256,
right_wrist 8/256. Built with build_cache_yam.py (spike > 3x neighbor mean in
=50% of 12 synthetic frames x 18 VLM layers).
Load: torch.load("cache_yam.pt")["masks"] # {cam_idx: [16,16] bool}