imported>Willy1018-bot：BOTR：批量替換所有lang-xx模板為langx模板

2025-06-25T12:20:33Z

BOTR：批量替換所有lang-xx模板為langx模板

新页面

{{noteTA|G1=IT}}
{{Infobox software
| title = 生成型预训练变换模型 2<br />Generative Pre-trained Transformer 2（GPT-2）
| logo =
| screenshot = File:GPT2-talks-about-GPT2.png
| screenshot_size =
| caption = GPT-2 使用 [[Hugging Face]] Write With Transformer 网站完成的文本，提示文字来自[[維基百科]]（初始提示后所有突出显示的文本都是从第一个建议的完成机器生成的，没有进一步编辑）
| author = [[OpenAI]]
| replaces = [[GPT-1]]
| replaced_by = [[GPT-3]]
| developer =
| released = {{Start date and age|2019|2|14}}
| programming language =
| operating system =
| genre = {{ indented plainlist |
*[[大型语言模型]]
*[[基于转换器的生成式预训练模型]]
}}
| license =
| repo = https://github.com/openai/gpt-2
| website = {{url|https://openai.com/blog/gpt-2-1-5b-release/}}
}}

'''生成式预训练变换模型2'''（{{langx|en|Generative Pre-trained Transformer 2}}，简称 '''GPT-2'''）是[[OpenAI]]于2019年2月创建的[[开源]][[人工智能]]。<ref name="voxxy2">{{Cite web |title=A poetry-writing AI has just been unveiled. It's ... pretty good. |url=https://www.vox.com/2019/5/15/18623134/openai-language-ai-gpt2-poetry-try-it |website=[[Vox Media|Vox]] |date=15 May 2019 |last=Piper |first=Kelsey |url-status=live |archive-url=https://web.archive.org/web/20201107235346/https://www.vox.com/2019/5/15/18623134/openai-language-ai-gpt2-poetry-try-it |archive-date=7 November 2020 |access-date=19 December 2020 |quote=}}</ref><ref name="vb">{{Cite web |title=OpenAI releases curtailed version of GPT-2 language model |url=https://venturebeat.com/2019/08/20/openai-releases-curtailed-version-of-gpt-2-language-model/ |website= |date=20 August 2019 |last=Johnson |first=Khari |url-status=live |archive-url=https://web.archive.org/web/20201218130530/https://venturebeat.com/2019/08/20/openai-releases-curtailed-version-of-gpt-2-language-model/ |archive-date=18 December 2020 |access-date=19 December 2020 |publisher=[[VentureBeat]] |quote=}}</ref><ref name="verge2">{{Cite web |title=OpenAI has published the text-generating AI it said was too dangerous to share |url=https://www.theverge.com/2019/11/7/20953040/openai-text-generation-ai-gpt-2-full-model-release-1-5b-parameters |website=[[The Verge]] |date=7 November 2019 |last=Vincent |first=James |url-status=live |archive-url=https://web.archive.org/web/20200611054114/https://www.theverge.com/2019/11/7/20953040/openai-text-generation-ai-gpt-2-full-model-release-1-5b-parameters |archive-date=11 June 2020 |access-date=19 December 2020 |quote=}}</ref><ref name="openai">{{Cite web |title=Better Language Models and Their Implications |url=https://openai.com/blog/better-language-models/ |website= |date=14 February 2019 |last= |first= |url-status=live |archive-url=https://web.archive.org/web/20201219132206/https://openai.com/blog/better-language-models/ |archive-date=19 December 2020 |access-date=19 December 2020 |publisher=[[OpenAI]] |quote=}}</ref>GPT-2能够[[机器翻译|翻译]]文本、[[問答系統|回答问题]]、[[自动总结|总结]]段落，<ref name="badpaper">{{Cite arXiv |arxiv=2006.05477 |last=Hegde |title=Unsupervised Paraphrase Generation using Pre-trained Language Models}}</ref>并[[自然语言生成|生成文本输出]]。虽然其输出内容有时[[图灵测试|与人类相似]]，<ref name="tds2">{{Cite web |title=Too big to deploy: How GPT-2 is breaking servers |url=https://towardsdatascience.com/too-big-to-deploy-how-gpt-2-is-breaking-production-63ab29f0897c |website=Towards Data Science |date=31 January 2020 |last=Kaiser |first=Caleb |url-status=live |archive-url=https://web.archive.org/web/20200215145640/https://towardsdatascience.com/too-big-to-deploy-how-gpt-2-is-breaking-production-63ab29f0897c |archive-date=15 February 2020 |access-date=27 February 2021 |publisher= |quote=}}</ref>但在生成长段落时输出内容可能会变得重复或无意义。<ref name="guardian">{{Cite web |title=New AI fake text generator may be too dangerous to release, say creators |url=https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction |website=[[The Guardian]] |date=14 February 2019 |last=Hern |first=Alex |url-status=live |archive-url=https://web.archive.org/web/20190214173112/https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction |archive-date=14 February 2019 |access-date=19 December 2020 |quote=}}</ref>GPT-2 是一个[[通用人工智慧|通用学习器]]，没有经过专门训练来执行任何特定的任务，<ref name="badpaper" /><ref name="gpt2paper">{{Cite journal |last=Radford |first=Alec |last2=Wu |first2=Jeffrey |last3=Child |first3=Rewon |last4=Luan |first4=David |last5=Amodei |first5=Dario |last6=Sutskever |first6=Ilua |title=Language models are unsupervised multitask learners |url=https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf |url-status=live |journal= |date=14 February 2019 |volume=1 |issue=8 |archive-url=https://web.archive.org/web/20210206183945/https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf |archive-date=6 February 2021 |access-date=19 December 2020 |quote=}}</ref>并且是作为 OpenAI 2018 GPT 模型的“直接扩展”而创建的，<ref name="gpt1paper">{{Cite web |title=Improving Language Understanding by Generative Pre-Training |url=https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf |date=11 June 2018 |last=Radford |first=Alec |last2=Narasimhan |first2=Karthik |last3=Salimans |first3=Tim |last4=Sutskever |first4=Ilya |url-status=live |archive-url=https://web.archive.org/web/20210126024542/https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf |archive-date=26 January 2021 |access-date=23 January 2021 |publisher=[[OpenAI]] |page=12}}</ref>其参数数量和训练数据集的大小均增加了十倍。<ref name="openai" />

==参考資料==
{{reflist|2}}

{{OpenAI}}
{{Differentiable computing}}

[[Category:OpenAI]]
[[Category:使用MIT許可證的軟體]]
[[Category:大型语言模型]]
[[Category:基于转换器的生成式预训练模型]]

GPT-2 - 版本历史

imported>Willy1018-bot：​BOTR：批量替換所有lang-xx模板為langx模板

imported>Willy1018-bot：BOTR：批量替換所有lang-xx模板為langx模板