使用测试版本
1
2
3
4
| torch 2.2.0
Python 3.10.16
CUDA 12.4
H200
|
500 trials by default (10 tasks x 50 episodes each)
| libero | spatial | object | goal | long | avg |
|---|
| H200 | 90.2% | 98.8% | 97.8% | 92.6% | 94.9% |
| H200(align) | 98.4% | 99.8% | 97.6% | 93.2% | 97.25% |
| 论文报告 | 99.4% | 99.6% | 98.8% | 96.0% | 98.5% |
libero_spatial
1
| python experiments/robot/libero/run_libero_eval.py --pretrained_checkpoint ckpts/spatial-forcing-7b-finetuned-libero-spatial --task_suite_name libero_spatial
|
1
2
3
4
| INFO | >> Final results: run_libero_eval.py:222
INFO | >> Total episodes: 500 run_libero_eval.py:222
INFO | >> Total successes: 451 run_libero_eval.py:222
INFO | >> Overall success rate: 0.9020 (90.2%) run_libero_eval.py:222
|
libero_object
1
| python experiments/robot/libero/run_libero_eval.py --pretrained_checkpoint ckpts/spatial-forcing-7b-finetuned-libero-object --task_suite_name libero_object
|
1
2
3
4
| INFO | >> Final results: run_libero_eval.py:222
INFO | >> Total episodes: 500 run_libero_eval.py:222
INFO | >> Total successes: 494 run_libero_eval.py:222
INFO | >> Overall success rate: 0.9880 (98.8%)
|
libero_goal
1
| ython experiments/robot/libero/run_libero_eval.py --pretrained_checkpoint ckpts/spatial-forcing-7b-finetuned-libero-goal --task_suite_name libero_goal
|
1
2
3
4
| INFO | >> Final results: run_libero_eval.py:222
INFO | >> Total episodes: 500 run_libero_eval.py:222
INFO | >> Total successes: 489 run_libero_eval.py:222
INFO | >> Overall success rate: 0.9780 (97.8%)
|
libero_10 / libero-long
1
| python experiments/robot/libero/run_libero_eval.py --pretrained_checkpoint ckpts/spatial-forcing-7b-finetuned-libero-10 --task_suite_name libero_10
|
1
2
3
4
| INFO | >> Final results: run_libero_eval.py:222
INFO | >> Total episodes: 500 run_libero_eval.py:222
INFO | >> Total successes: 463 run_libero_eval.py:222
INFO | >> Overall success rate: 0.9260 (92.6%)
|