bridgequant-so101-7m-kd

BridgeQuant 7M student distilled from the 76M SO101 teacher (twanghcmut/bridgequant-so101-76m) on the three SO101 follower-arm tasks (159 episodes, 30 fps, cameras top+wrist, 6-D absolute joint targets). KD: role-group-pooled latent loss (1-cos + Huber, 0.1) + L1 on the clean-action estimate (0.5, annealed to 0 over the last 30%).

Files: model.safetensors, norm_stats.json, bridgequant_vla.yaml / config.json / metadata.json, train_config.yaml (the teacher paths inside are only needed for training; inference ignores them).

Code: https://github.com/VLAOpt/bridgequant-async — scripts/so101_policy.py, guide in docs/so101_deploy.md.