Instructions to use celestial4SR/statebackdoor-openvla-libero-object with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use celestial4SR/statebackdoor-openvla-libero-object with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="celestial4SR/statebackdoor-openvla-libero-object", trust_remote_code=True)# Load model directly from transformers import AutoModelForVision2Seq model = AutoModelForVision2Seq.from_pretrained("celestial4SR/statebackdoor-openvla-libero-object", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
StateBackdoor OpenVLA LIBERO-Object Checkpoint
This private repository contains the StateBackdoor OpenVLA checkpoint used for SecurityCuff benchmark integration.
- Base model:
/home/srpan/WorkSpace/data/openvla-7b - Benchmark environment: GoBA fork of LIBERO,
libero_object - Trigger:
/home/srpan/WorkSpace/data/state_backdoor/triggers/policy_start_fixed_nn_v2/s_trig.json - Training run:
statebackdoor_poison_basket_lenmatch_v3_from7000_add8k_3g_b8a5 - Checkpoint path at upload time:
/home/srpan/WorkSpace/data/state_backdoor/runs/statebackdoor_poison_basket_lenmatch_v3_from7000_add8k_3g_b8a5/sb_poison_v3_7000+libero_object_no_noops+b40+lr-0.0005+lora-r32+dropout-0.0--state_backdoor_statebackdoor_poison_basket_lenmatch_v3_from7000_add8k_3g_b8a5--image_aug--8000_chkpt - Approximate checkpoint size:
15G
Evaluation summary under GoBA-equivalent env_reset + wait10 protocol:
- Clean SR:
0.75 - Triggered original-task SR:
0.0 - Target-alignment ASR:
None
Use only for authorized research on backdoor detection and SecurityCuff benchmark reproduction.
- Downloads last month
- 25