Modelpublic

s1ghhh/VLADrop_pi05_LIBERO_Drop25VisionBlock_Endpoints

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes

Model Card

VLADrop-pi05-LIBERO-vision-keep2

Checkpoint for Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?.

DTR (Drop-Then-Recovery) removes transformer blocks from a pretrained VLA model and recovery-fine-tunes the smaller dense model. Code: https://github.com/s1ghhh/VLADrop

This checkpoint


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

Usage

This is an openpi-format pi0.5 checkpoint (PyTorch). Use with the VLADrop fork: https://github.com/s1ghhh/VLADrop

bash

python scripts/serve_policy_batch_drop.py \
    --config pi05_libero_dropped \
    --dir <this_repo_local_path> \
    --port 8000

Important: the drop lists are NOT stored inside the checkpoint. Pass the exact llm_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.

Citation

bibtex

@article{sun2026vladrop,
  title={Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?},
  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},
  journal={arXiv preprint arXiv:2606.27755},
  year={2026}
}