The success rate (%) of BitVLA and the baselines on LIBERO simulation environment

Models Size Memory Usage↓ Spatial Object Goal Long Avg.
w/ Robotics pre-training
OpenVLA 7.5B 15.1GB (10.79×) 84.7 88.4 79.2 53.7 76.5
SpatialVLA 4.2B 8.5GB (6.07×) 88.2 89.9 78.6 55.5 78.1
CoT-VLA 8.0B 16.2GB (11.57×) 87.5 91.6 87.6 69.0 81.1
NORA-Long 3.8B 7.5GB (5.36×) 92.2 95.4 89.4 74.6 87.9
π₀ 3.5B 7.0GB (5.00×) 96.8 98.8 95.8 85.2 94.2
OpenVLA-OFT 7.7B 15.4GB (11.00×) 97.6 98.4 97.9 94.5 97.1
w/o Robotics pre-training
OpenVLA-OFT 7.7B 15.4GB (11.00×) 94.3 95.2 91.7 86.5 91.9
BitVLA 3.0B 1.4GB (1.00×) 97.4 99.6 94.4 87.6 94.8
Downloads last month
5
Safetensors
Model size
2.82B params
Tensor type
BF16
·
Video Preview
loading

Model tree for hongyuw/ft-bitvla-bitsiglipL-224px-libero_object-bf16

Finetuned
(4)
this model

Dataset used to train hongyuw/ft-bitvla-bitsiglipL-224px-libero_object-bf16

Collection including hongyuw/ft-bitvla-bitsiglipL-224px-libero_object-bf16