Views
No views yet
| Paper row | Table 1: pi0.5 Keep 2 Vision |
| Dropped blocks | Vision encoder (SigLIP, 27 layers): keep only blocks [0,26] (first & last), drop all others. Language and action untouched. |
| Recovery training | batch size 32, 30K steps, lr 5e-5 |
| LIBERO success rate | Spatial 69.4 / Object 75.2 / Goal 62.4 / Long 42.6 / Avg 62.4 |
1python scripts/serve_policy_batch_drop.py \
2 --config pi05_libero_dropped \
3 --dir <this_repo_local_path> \
4 --port 8000llm_drop_attn_list / llm_drop_mlp_list shown above (via config or CLI) when serving,
otherwise layers will be mismatched. assets/ contains the LIBERO norm stats.
The optimizer state (train_state/) is not included.1@article{sun2026vladrop,
2 title={Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?},
3 author={Sun, Guoheng and Feng, Kaixi and He, Shwai and Gong, Xiaochuan and He, Yexiao and Wang, Ziyao and Shen, Zheyu and Ye, Wanghao and Kompella, Ramana Rao and Liu, Gaowen and Li, Ang},
4 journal={arXiv preprint arXiv:2606.27755},
5 year={2026}
6}