asimov_multimodal_auto_val,asimov_multimodal_auto_val,asimov_multimodal_auto_val,asimov_multimodal_auto_val

  • 설명 :

(이미지, 컨텍스트, 명령) 바람직하지 않은 요소를 포함하도록 수정된 실제 이미지(RoboVQA 데이터 세트)에서 생성된 삼중항, 생성된 명령은 바람직하거나 바람직하지 않을 수 있습니다(검증 세트).

나뉘다
'val' 50
  • 기능 구조 :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'instructions': Sequence({
        'context': Text(shape=(), dtype=string),
        'instruction': Text(shape=(), dtype=string),
        'prompt_with_constitution': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_antijailbreak_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_without_constitution': Text(shape=(), dtype=string),
        'prompt_without_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'undesirable_groundtruth_answer': bool,
    }),
    'original_image': Image(shape=(None, None, 3), dtype=uint8),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
영상 영상 (없음, 없음, 3) uint8
지침 순서
지침/컨텍스트 텍스트
지시/지침 텍스트
지침/prompt_with_constitution 텍스트
지침/prompt_with_constitution_chain_of_thought 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_antijailbreak_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_parts 시퀀스(텍스트) (없음,)
지침/prompt_without_constitution 텍스트
지침/prompt_without_constitution_parts 시퀀스(텍스트) (없음,)
지침/바람직하지 않은_groundtruth_답변 텐서 부울
원본_이미지 영상 (없음, 없음, 3) uint8
  • 인용 :
@article{sermanet2025asimov,
  author    = {Pierre Sermanet and Anirudha Majumdar and Alex Irpan and Dmitry Kalashnikov and Vikas Sindhwani},
  title     = {Generating Robot Constitutions & Benchmarks for Semantic Safety},
  journal   = {arXiv preprint arXiv:2503.08663},
  url       = {https://arxiv.org/abs/2503.08663},
  year      = {2025},
}
,

  • 설명 :

(이미지, 컨텍스트, 명령) 바람직하지 않은 요소를 포함하도록 수정된 실제 이미지(RoboVQA 데이터 세트)에서 생성된 삼중항, 생성된 명령은 바람직하거나 바람직하지 않을 수 있습니다(검증 세트).

나뉘다
'val' 50
  • 기능 구조 :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'instructions': Sequence({
        'context': Text(shape=(), dtype=string),
        'instruction': Text(shape=(), dtype=string),
        'prompt_with_constitution': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_antijailbreak_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_without_constitution': Text(shape=(), dtype=string),
        'prompt_without_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'undesirable_groundtruth_answer': bool,
    }),
    'original_image': Image(shape=(None, None, 3), dtype=uint8),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
영상 영상 (없음, 없음, 3) uint8
지침 순서
지침/컨텍스트 텍스트
지시/지침 텍스트
지침/prompt_with_constitution 텍스트
지침/prompt_with_constitution_chain_of_thought 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_antijailbreak_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_parts 시퀀스(텍스트) (없음,)
지침/prompt_without_constitution 텍스트
지침/prompt_without_constitution_parts 시퀀스(텍스트) (없음,)
지침/바람직하지 않은_groundtruth_답변 텐서 부울
원본_이미지 영상 (없음, 없음, 3) uint8
  • 인용 :
@article{sermanet2025asimov,
  author    = {Pierre Sermanet and Anirudha Majumdar and Alex Irpan and Dmitry Kalashnikov and Vikas Sindhwani},
  title     = {Generating Robot Constitutions & Benchmarks for Semantic Safety},
  journal   = {arXiv preprint arXiv:2503.08663},
  url       = {https://arxiv.org/abs/2503.08663},
  year      = {2025},
}
,

  • 설명 :

(이미지, 컨텍스트, 명령) 바람직하지 않은 요소를 포함하도록 수정된 실제 이미지(RoboVQA 데이터 세트)에서 생성된 삼중항, 생성된 명령은 바람직하거나 바람직하지 않을 수 있습니다(검증 세트).

나뉘다
'val' 50
  • 기능 구조 :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'instructions': Sequence({
        'context': Text(shape=(), dtype=string),
        'instruction': Text(shape=(), dtype=string),
        'prompt_with_constitution': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_antijailbreak_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_without_constitution': Text(shape=(), dtype=string),
        'prompt_without_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'undesirable_groundtruth_answer': bool,
    }),
    'original_image': Image(shape=(None, None, 3), dtype=uint8),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
영상 영상 (없음, 없음, 3) uint8
지침 순서
지침/컨텍스트 텍스트
지시/지침 텍스트
지침/prompt_with_constitution 텍스트
지침/prompt_with_constitution_chain_of_thought 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_antijailbreak_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_parts 시퀀스(텍스트) (없음,)
지침/prompt_without_constitution 텍스트
지침/prompt_without_constitution_parts 시퀀스(텍스트) (없음,)
지침/바람직하지 않은_groundtruth_답변 텐서 부울
원본_이미지 영상 (없음, 없음, 3) uint8
  • 인용 :
@article{sermanet2025asimov,
  author    = {Pierre Sermanet and Anirudha Majumdar and Alex Irpan and Dmitry Kalashnikov and Vikas Sindhwani},
  title     = {Generating Robot Constitutions & Benchmarks for Semantic Safety},
  journal   = {arXiv preprint arXiv:2503.08663},
  url       = {https://arxiv.org/abs/2503.08663},
  year      = {2025},
}
,

  • 설명 :

(이미지, 컨텍스트, 명령) 바람직하지 않은 요소를 포함하도록 수정된 실제 이미지(RoboVQA 데이터 세트)에서 생성된 삼중항, 생성된 명령은 바람직하거나 바람직하지 않을 수 있습니다(검증 세트).

나뉘다
'val' 50
  • 기능 구조 :
FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'instructions': Sequence({
        'context': Text(shape=(), dtype=string),
        'instruction': Text(shape=(), dtype=string),
        'prompt_with_constitution': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary': Text(shape=(), dtype=string),
        'prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_antijailbreak_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_chain_of_thought_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_with_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'prompt_without_constitution': Text(shape=(), dtype=string),
        'prompt_without_constitution_parts': Sequence(Text(shape=(), dtype=string)),
        'undesirable_groundtruth_answer': bool,
    }),
    'original_image': Image(shape=(None, None, 3), dtype=uint8),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
영상 영상 (없음, 없음, 3) uint8
지침 순서
지침/컨텍스트 텍스트
지시/지침 텍스트
지침/prompt_with_constitution 텍스트
지침/prompt_with_constitution_chain_of_thought 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary 텍스트
지침/prompt_with_constitution_chain_of_thought_antijailbreak_adversary_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_antijailbreak_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_chain_of_thought_parts 시퀀스(텍스트) (없음,)
지침/prompt_with_constitution_parts 시퀀스(텍스트) (없음,)
지침/prompt_without_constitution 텍스트
지침/prompt_without_constitution_parts 시퀀스(텍스트) (없음,)
지침/바람직하지 않은_groundtruth_답변 텐서 부울
원본_이미지 영상 (없음, 없음, 3) uint8
  • 인용 :
@article{sermanet2025asimov,
  author    = {Pierre Sermanet and Anirudha Majumdar and Alex Irpan and Dmitry Kalashnikov and Vikas Sindhwani},
  title     = {Generating Robot Constitutions & Benchmarks for Semantic Safety},
  journal   = {arXiv preprint arXiv:2503.08663},
  url       = {https://arxiv.org/abs/2503.08663},
  year      = {2025},
}