https://zh.lolipedia.org/index.php?action=history&feed=atom&title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B 2026-04-07T17:15:37Z 本wiki上该页面的版本历史 MediaWiki 1.44.0 https://zh.lolipedia.org/index.php?title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&diff=17&oldid=prev 2026-04-07T00:37:40Z

<p></p> <table style="background-color: #fff; color: #202122;" data-mw="interface"> <col class="diff-marker" /> <col class="diff-content" /> <col class="diff-marker" /> <col class="diff-content" /> <tr class="diff-title" lang="zh"> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">←上一版本</td> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">2026年4月7日 (二) 08:37的版本</td> </tr><tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l1">第1行：</td> <td colspan="2" class="diff-lineno">第1行：</td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>'''大语言模型'''（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。</div></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>'''大语言模型'''（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言<ins style="font-weight: bold; text-decoration: none;">的统计</ins>规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。</div></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>== 核心架构：Transformer ==</div></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>== 核心架构：Transformer ==</div></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>现代LLM大多基于'''Transformer'''架构。该<del style="font-weight: bold; text-decoration: none;">的统计</del></div></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>现代LLM大多基于'''Transformer'''架构。该<ins style="font-weight: bold; text-decoration: none;">架</ins>构<ins style="font-weight: bold; text-decoration: none;">引入了注意力机制（Attention Mechanism），使模型</ins>能够<ins style="font-weight: bold; text-decoration: none;">捕捉长距离</ins>的<ins style="font-weight: bold; text-decoration: none;">语义依赖</ins>。</div></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">的统计规律和语义结</del>构<del style="font-weight: bold; text-decoration: none;">。目前的LLM已经</del>能够<del style="font-weight: bold; text-decoration: none;">实现流畅</del>的<del style="font-weight: bold; text-decoration: none;">多轮对话、代码编写以及复杂的逻辑推理</del>。</div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>== <del style="font-weight: bold; text-decoration: none;">核心架</del>构<del style="font-weight: bold; text-decoration: none;">：Transformer </del>==</div></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div>== <ins style="font-weight: bold; text-decoration: none;">训练过程 ==</ins></div></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div>现<del style="font-weight: bold; text-decoration: none;">代LLM大</del>多<del style="font-weight: bold; text-decoration: none;">基于</del>'''<del style="font-weight: bold; text-decoration: none;">Transformer</del>'''<del style="font-weight: bold; text-decoration: none;">架构</del></div></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">LLM的</ins>构<ins style="font-weight: bold; text-decoration: none;">建通常包含以下两个关键阶段：</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"># '''预训练（Pre-training）'''：模型在大规模的互联网文本语料库上进行无监督学习。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"># '''指令微调（Instruction Tuning）'''：通过人类反馈强化学习等技术，教会模型如何准确理解并执行用户的指令。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div> </div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 局限性：幻觉 </ins>==</div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">尽管LLM表</ins>现<ins style="font-weight: bold; text-decoration: none;">卓越，但仍存在幻觉现象，即模型可能会生成看似合理但完全错误的事实。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div> </div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 应用领域 ==</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">LLM目前已广泛应用于多个前沿领域：</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* '''智能交互'''：作为虚拟助手提供情感陪伴与信息查询。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* '''辅助开发'''：帮助工程师快速生成代码片段。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">* '''内容创作'''：协助进行</ins>多<ins style="font-weight: bold; text-decoration: none;">语言翻译、文章撰写。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div> </div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 参考资料 ==</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"># [https://arxiv.org/abs/1706.03762 Vaswani, A., et al. (2017). ''Attention Is All You Need''].</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"># [https://openai.com/research/gpt-4 OpenAI. (2023). ''GPT-4 Technical Report</ins>''<ins style="font-weight: bold; text-decoration: none;">].</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"># [https://stanford.io/3lgdH2N Bommasani, R., et al. (2021). </ins>''<ins style="font-weight: bold; text-decoration: none;">On the Opportunities and Risks of Foundation Models</ins>''<ins style="font-weight: bold; text-decoration: none;">].</ins></div></td></tr> <!-- diff cache key zhwiki:diff:1.41:old-16:rev-17:php=table --> </table>

Yuqing https://zh.lolipedia.org/index.php?title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&diff=16&oldid=prev 2026-04-07T00:07:00Z

<p></p> <table style="background-color: #fff; color: #202122;" data-mw="interface"> <col class="diff-marker" /> <col class="diff-content" /> <col class="diff-marker" /> <col class="diff-content" /> <tr class="diff-title" lang="zh"> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">←上一版本</td> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">2026年4月7日 (二) 08:07的版本</td> </tr><tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l3">第3行：</td> <td colspan="2" class="diff-lineno">第3行：</td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>== 核心架构：Transformer ==</div></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>== 核心架构：Transformer ==</div></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>现代LLM大多基于&#039;&#039;&#039;Transformer&#039;&#039;&#039;架构。该的统计</div></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><div>现代LLM大多基于&#039;&#039;&#039;Transformer&#039;&#039;&#039;架构。该的统计</div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">的统计规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;"></ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 核心架构：Transformer ==</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">现代LLM大多基于'''Transformer'''架构</ins></div></td></tr> <!-- diff cache key zhwiki:diff:1.41:old-15:rev-16:php=table --> </table>

Yuqing https://zh.lolipedia.org/index.php?title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&diff=15&oldid=prev 2026-04-07T00:02:46Z

<p></p> <table style="background-color: #fff; color: #202122;" data-mw="interface"> <col class="diff-marker" /> <col class="diff-content" /> <col class="diff-marker" /> <col class="diff-content" /> <tr class="diff-title" lang="zh"> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">←上一版本</td> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">2026年4月7日 (二) 08:02的版本</td> </tr><tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l1">第1行：</td> <td colspan="2" class="diff-lineno">第1行：</td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">'''大语言模型'''（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。</ins></div></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">== 核心架构：Transformer ==</ins></div></td></tr> <tr><td colspan="2" class="diff-side-deleted"></td><td class="diff-marker" data-marker="+"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;"><div><ins style="font-weight: bold; text-decoration: none;">现代LLM大多基于'''Transformer'''架构。该的统计</ins></div></td></tr> <!-- diff cache key zhwiki:diff:1.41:old-14:rev-15:php=table --> </table>

Yuqing https://zh.lolipedia.org/index.php?title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&diff=14&oldid=prev 2026-04-06T23:49:17Z

<p>清空全部内容</p> <table style="background-color: #fff; color: #202122;" data-mw="interface"> <col class="diff-marker" /> <col class="diff-content" /> <col class="diff-marker" /> <col class="diff-content" /> <tr class="diff-title" lang="zh"> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">←上一版本</td> <td colspan="2" style="background-color: #fff; color: #202122; text-align: center;">2026年4月7日 (二) 07:49的版本</td> </tr><tr><td colspan="2" class="diff-lineno" id="mw-diff-left-l1">第1行：</td> <td colspan="2" class="diff-lineno">第1行：</td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">'''大语言模型'''（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言的统计规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td><td class="diff-marker"></td><td style="background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;"><br></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">== 核心架构：Transformer ==</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">现代LLM大多基于'''Transformer'''架构。该架构引入了注意力机制（Attention Mechanism），使模型能够捕捉长距离的语义依赖。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">== 训练过程 ==</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">LLM的构建通常包含以下两个关键阶段：</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"># '''预训练（Pre-training）'''：模型在大规模的互联网文本语料库上进行无监督学习。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"># '''指令微调（Instruction Tuning）'''：通过人类反馈强化学习等技术，教会模型如何准确理解并执行用户的指令。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">== 局限性：幻觉 ==</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">尽管LLM表现卓越，但仍存在幻觉现象，即模型可能会生成看似合理但完全错误的事实。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">== 应用领域 ==</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">LLM目前已广泛应用于多个前沿领域：</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">* '''智能交互'''：作为虚拟助手提供情感陪伴与信息查询。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">* '''辅助开发'''：帮助工程师快速生成代码片段。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">* '''内容创作'''：协助进行多语言翻译、文章撰写。</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;">== 参考资料 ==</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"># Vaswani, A., et al. (2017). ''Attention Is All You Need''.</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"># OpenAI. (2023). ''GPT-4 Technical Report''.</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <tr><td class="diff-marker" data-marker="−"></td><td style="color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;"><div><del style="font-weight: bold; text-decoration: none;"># Bommasani, R., et al. (2021). ''On the Opportunities and Risks of Foundation Models''.</del></div></td><td colspan="2" class="diff-side-added"></td></tr> <!-- diff cache key zhwiki:diff:1.41:old-13:rev-14:php=table --> </table>

Yuqing https://zh.lolipedia.org/index.php?title=%E5%A4%A7%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&diff=13&oldid=prev 2026-04-06T23:44:03Z

<p>创建页面，内容为“&#039;&#039;&#039;大语言模型&#039;&#039;&#039;（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言的统计规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。 == 核心架构：Transformer == 现代LLM大多基于&#039;&#039;&#039;Transformer&#039;&#039;&#039;架构。该架构引入了注意力…”</p> <p><b>新页面</b></p><div>&#039;&#039;&#039;大语言模型&#039;&#039;&#039;（Large Language Model，简称LLM）是一种基于深度学习的人工智能模型，专门用于处理、理解和生成人类语言。它通过在海量的文本数据上进行训练，学会了语言的统计规律和语义结构。目前的LLM已经能够实现流畅的多轮对话、代码编写以及复杂的逻辑推理。<br /> <br /> == 核心架构：Transformer ==<br /> 现代LLM大多基于&#039;&#039;&#039;Transformer&#039;&#039;&#039;架构。该架构引入了注意力机制（Attention Mechanism），使模型能够捕捉长距离的语义依赖。<br /> <br /> == 训练过程 ==<br /> LLM的构建通常包含以下两个关键阶段：<br /> # &#039;&#039;&#039;预训练（Pre-training）&#039;&#039;&#039;：模型在大规模的互联网文本语料库上进行无监督学习。<br /> # &#039;&#039;&#039;指令微调（Instruction Tuning）&#039;&#039;&#039;：通过人类反馈强化学习等技术，教会模型如何准确理解并执行用户的指令。<br /> <br /> == 局限性：幻觉 ==<br /> 尽管LLM表现卓越，但仍存在幻觉现象，即模型可能会生成看似合理但完全错误的事实。<br /> <br /> == 应用领域 ==<br /> LLM目前已广泛应用于多个前沿领域：<br /> * &#039;&#039;&#039;智能交互&#039;&#039;&#039;：作为虚拟助手提供情感陪伴与信息查询。<br /> * &#039;&#039;&#039;辅助开发&#039;&#039;&#039;：帮助工程师快速生成代码片段。<br /> * &#039;&#039;&#039;内容创作&#039;&#039;&#039;：协助进行多语言翻译、文章撰写。<br /> <br /> == 参考资料 ==<br /> # Vaswani, A., et al. (2017). &#039;&#039;Attention Is All You Need&#039;&#039;.<br /> # OpenAI. (2023). &#039;&#039;GPT-4 Technical Report&#039;&#039;.<br /> # Bommasani, R., et al. (2021). &#039;&#039;On the Opportunities and Risks of Foundation Models&#039;&#039;.</div>

Yuqing