वास्तविक_विषाक्तता_संकेत

  • विवरण :

RealToxicityPrompts: भाषा मॉडल में तंत्रिका विषाक्त अध: पतन का मूल्यांकन

RealToxicityPrompts अलग-अलग विषाक्तता के अंग्रेजी संकेतों (विशेष रूप से वाक्य शुरुआत) का एक संग्रह है। ये संकेत एक ऑटोरेग्रेसिव भाषा मॉडल (यानी GPT-2) के संदर्भ के रूप में दिए जाने हैं और पूर्णता उत्पन्न करने के लिए उपयोग किए जाते हैं।

अधिक विवरण मूल पेपर में प्रस्तुत किए गए हैं।

विभाजित करना उदाहरण
'train' 99,442
  • फ़ीचर संरचना :
FeaturesDict({
    'begin': int32,
    'challenging': bool,
    'continuation': FeaturesDict({
        'flirtation': float32,
        'identity_attack': float32,
        'insult': float32,
        'profanity': float32,
        'severe_toxicity': float32,
        'sexually_explicit': float32,
        'text': Text(shape=(), dtype=string),
        'threat': float32,
        'toxicity': float32,
    }),
    'end': int32,
    'filename': Text(shape=(), dtype=string),
    'prompt': FeaturesDict({
        'flirtation': float32,
        'identity_attack': float32,
        'insult': float32,
        'profanity': float32,
        'severe_toxicity': float32,
        'sexually_explicit': float32,
        'text': Text(shape=(), dtype=string),
        'threat': float32,
        'toxicity': float32,
    }),
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीप्रकार विवरण
फीचर्सडिक्ट
शुरू टेन्सर int32
चुनौतीपूर्ण टेन्सर बूल
विस्तार फीचर्सडिक्ट
निरंतरता/इश्कबाज़ी टेन्सर फ्लोट32
निरंतरता/पहचान_हमला टेन्सर फ्लोट32
निरंतरता/अपमान टेन्सर फ्लोट32
निरंतरता/अपवित्रता टेन्सर फ्लोट32
निरंतरता/गंभीर_विषाक्तता टेन्सर फ्लोट32
निरंतरता/यौन_स्पष्ट टेन्सर फ्लोट32
निरंतरता/पाठ मूलपाठ डोरी
निरंतरता/धमकी टेन्सर फ्लोट32
निरंतरता/विषाक्तता टेन्सर फ्लोट32
अंत टेन्सर int32
फ़ाइल का नाम मूलपाठ डोरी
तत्पर फीचर्सडिक्ट
संकेत/इश्कबाज़ी टेन्सर फ्लोट32
संकेत/पहचान_हमला टेन्सर फ्लोट32
संकेत/अपमान टेन्सर फ्लोट32
संकेत/अपवित्रता टेन्सर फ्लोट32
शीघ्र/गंभीर_विषाक्तता टेन्सर फ्लोट32
शीघ्र/यौन_स्पष्ट टेन्सर फ्लोट32
शीघ्र/पाठ मूलपाठ डोरी
संकेत/धमकी टेन्सर फ्लोट32
शीघ्र/विषाक्तता टेन्सर फ्लोट32
  • उद्धरण :
@article{gehman2020realtoxicityprompts,
  title={Realtoxicityprompts: Evaluating neural toxic degeneration in language models},
  author={Gehman, Samuel and Gururangan, Suchin and Sap, Maarten and Choi, Yejin and Smith, Noah A},
  journal={arXiv preprint arXiv:2009.11462},
  year={2020}
}