Language-conditioned robotic manipulation on LIBERO
97.6Spatial Success RateOpenVLA-OFT (FP Model)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OpenVLA-OFT (FP Model)Model=OpenVLA-OFT, Quantization Setting=FP Model2026.06 | 97.6 | 98.4 | 97.9 | 94.5 | 97.1 | — | 15.4 | |
| Mix-QVLAModel=OpenVLA-OFT, Quantization Setting=W8A162026.06 | 97.5 | 98.6 | 97.7 | 94.6 | 97.1 | 0 | 6.8 | |
| QVLAModel=OpenVLA-OFT, Quantization Setting=W8A162026.06 | 97.4 | 98.6 | 97.2 | 94.6 | 97 | -0.1 | 7.4 | |
| Mix-QVLAModel=OpenVLA-OFT, Quantization Setting=W4A162026.06 | 97.1 | 98.4 | 97.4 | 94.4 | 96.9 | -0.2 | 4.2 | |
| QVLAModel=OpenVLA-OFT, Quantization Setting=W4A162026.06 | 97 | 98.4 | 96.8 | 94.4 | 96.7 | -0.4 | 4.5 | |
| AWQModel=OpenVLA-OFT, Quantization Setting=W8A162026.06 | 95.2 | 96.8 | 95.4 | 93.1 | 95.1 | -2 | 8 | |
| AWQModel=OpenVLA-OFT, Quantization Setting=W4A162026.06 | 93 | 92.4 | 93.8 | 90.7 | 92.5 | -4.5 | 5.2 | |
| QVLAModel=OpenVLA, Quantization Setting=W8A162026.06 | 86.2 | 88.4 | 79.4 | 53.1 | 76.8 | 0.3 | 7.2 | |
| Mix-QVLAModel=OpenVLA, Quantization Setting=W8A162026.06 | 86 | 88.5 | 79.6 | 53.5 | 76.9 | 0.4 | 6.7 | |
| QVLAModel=OpenVLA, Quantization Setting=W4A162026.06 | 86 | 88.6 | 78.4 | 52.8 | 76.5 | 0 | 4.3 | |
| Mix-QVLAModel=OpenVLA, Quantization Setting=W4A162026.06 | 85.8 | 88.2 | 78.8 | 53.4 | 76.6 | 0.1 | 4.1 | |
| OpenVLA (FP Model)Model=OpenVLA, Quantization Setting=FP Model2026.06 | 84.7 | 88.4 | 79.2 | 53.7 | 76.5 | — | 15.2 | |
| AWQModel=OpenVLA, Quantization Setting=W8A162026.06 | 82.3 | 87.2 | 74.6 | 50.7 | 73.7 | -1.8 | 7.6 | |
| AWQModel=OpenVLA, Quantization Setting=W4A162026.06 | 80 | 81.2 | 74.6 | 47.2 | 70.8 | -4.7 | 5 |