flykhan
2536c937e3
翻译自英文原版 maths-cs-ai-compendium，共 20 章全部完成。 第01章 向量 | 第02章 矩阵 | 第03章 微积分 第04章 统计学 | 第05章 概率论 | 第06章 机器学习 第07章 计算语言学 | 第08章 计算机视觉 | 第09章 音频与语音 第10章 多模态学习 | 第11章 自主系统 | 第12章 图神经网络 第13章 计算与操作系统 | 第14章 数据结构与算法 第15章 生产级软件工程 | 第16章 SIMD与GPU编程 第17章 AI推理 | 第18章 ML系统设计 第19章 应用人工智能 | 第20章 前沿人工智能 翻译说明： - 所有数学公式 $...$ / $$...$$、代码块、图片引用完整保留 - mkdocs.yml 配置中文导航 + language: zh - README.md 已翻译为中文（兼 docs/index.md） - docs/ 目录包含指向各章文件的 symlink - 约 29,000 行中文内容，排除 .cache/ 构建缓存
103 lines
7.1 KiB
XML
103 lines
7.1 KiB
XML
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 650 300" width="650" height="300" font-family="Arial, sans-serif">
|
|
<!-- Title -->
|
|
<text x="325" y="22" text-anchor="middle" font-size="14" font-weight="bold" fill="#333">Choosing Between Continuous and Discrete Tokens</text>
|
|
|
|
<defs>
|
|
<marker id="dArr" markerWidth="8" markerHeight="6" refX="8" refY="3" orient="auto">
|
|
<path d="M0,0 L8,3 L0,6 Z" fill="#666"/>
|
|
</marker>
|
|
<marker id="dArrGreen" markerWidth="8" markerHeight="6" refX="8" refY="3" orient="auto">
|
|
<path d="M0,0 L8,3 L0,6 Z" fill="#27ae60"/>
|
|
</marker>
|
|
<marker id="dArrBlue" markerWidth="8" markerHeight="6" refX="8" refY="3" orient="auto">
|
|
<path d="M0,0 L8,3 L0,6 Z" fill="#3498db"/>
|
|
</marker>
|
|
<marker id="dArrOrange" markerWidth="8" markerHeight="6" refX="8" refY="3" orient="auto">
|
|
<path d="M0,0 L8,3 L0,6 Z" fill="#f39c12"/>
|
|
</marker>
|
|
</defs>
|
|
|
|
<!-- Start node -->
|
|
<rect x="220" y="42" width="210" height="40" rx="20" fill="#9b59b6" fill-opacity="0.12" stroke="#9b59b6" stroke-width="1.5"/>
|
|
<text x="325" y="66" text-anchor="middle" font-size="11" font-weight="bold" fill="#9b59b6">How will you generate?</text>
|
|
|
|
<!-- Left branch arrow -->
|
|
<path d="M260,82 L130,120" fill="none" stroke="#27ae60" stroke-width="1.5" marker-end="url(#dArrGreen)"/>
|
|
<text x="172" y="96" font-size="9" fill="#27ae60" font-weight="bold">Autoregressive</text>
|
|
<text x="170" y="108" font-size="8" fill="#666">(next-token)</text>
|
|
|
|
<!-- Right branch arrow -->
|
|
<path d="M390,82 L520,120" fill="none" stroke="#3498db" stroke-width="1.5" marker-end="url(#dArrBlue)"/>
|
|
<text x="462" y="96" font-size="9" fill="#3498db" font-weight="bold">Diffusion</text>
|
|
<text x="468" y="108" font-size="8" fill="#666">(iterative denoising)</text>
|
|
|
|
<!-- Middle branch arrow -->
|
|
<path d="M325,82 L325,120" fill="none" stroke="#f39c12" stroke-width="1.5" marker-end="url(#dArrOrange)"/>
|
|
<text x="358" y="106" font-size="9" fill="#f39c12" font-weight="bold">Hybrid</text>
|
|
|
|
<!-- Left: Discrete Tokens box -->
|
|
<rect x="30" y="128" width="200" height="80" rx="8" fill="#27ae60" fill-opacity="0.1" stroke="#27ae60" stroke-width="1.5"/>
|
|
<text x="130" y="152" text-anchor="middle" font-size="12" font-weight="bold" fill="#27ae60">Use Discrete Tokens</text>
|
|
<!-- Token grid illustration -->
|
|
<g transform="translate(60, 160)">
|
|
<rect x="0" y="0" width="22" height="18" rx="2" fill="#27ae60" fill-opacity="0.2" stroke="#27ae60" stroke-width="0.8"/>
|
|
<text x="11" y="12" text-anchor="middle" font-size="7" fill="#27ae60">42</text>
|
|
<rect x="25" y="0" width="22" height="18" rx="2" fill="#27ae60" fill-opacity="0.2" stroke="#27ae60" stroke-width="0.8"/>
|
|
<text x="36" y="12" text-anchor="middle" font-size="7" fill="#27ae60">7</text>
|
|
<rect x="50" y="0" width="22" height="18" rx="2" fill="#27ae60" fill-opacity="0.2" stroke="#27ae60" stroke-width="0.8"/>
|
|
<text x="61" y="12" text-anchor="middle" font-size="7" fill="#27ae60">156</text>
|
|
<rect x="75" y="0" width="22" height="18" rx="2" fill="#27ae60" fill-opacity="0.2" stroke="#27ae60" stroke-width="0.8"/>
|
|
<text x="86" y="12" text-anchor="middle" font-size="7" fill="#27ae60">89</text>
|
|
<rect x="100" y="0" width="22" height="18" rx="2" fill="#27ae60" fill-opacity="0.2" stroke="#27ae60" stroke-width="0.8"/>
|
|
<text x="111" y="12" text-anchor="middle" font-size="7" fill="#27ae60">3</text>
|
|
</g>
|
|
<text x="130" y="196" text-anchor="middle" font-size="9" fill="#666">VQ-VAE / VQ-GAN</text>
|
|
|
|
<!-- Left examples -->
|
|
<rect x="40" y="216" width="180" height="44" rx="6" fill="#27ae60" fill-opacity="0.05" stroke="#27ae60" stroke-width="0.8" stroke-dasharray="3,2"/>
|
|
<text x="130" y="232" text-anchor="middle" font-size="8" fill="#27ae60">DALL-E, Parti, LlamaGen,</text>
|
|
<text x="130" y="244" text-anchor="middle" font-size="8" fill="#27ae60">VideoGPT, MAGVIT</text>
|
|
|
|
<!-- Middle: Soft Quantisation -->
|
|
<rect x="245" y="128" width="160" height="80" rx="8" fill="#f39c12" fill-opacity="0.1" stroke="#f39c12" stroke-width="1.5"/>
|
|
<text x="325" y="150" text-anchor="middle" font-size="12" font-weight="bold" fill="#f39c12">Soft Quantisation</text>
|
|
<!-- Soft token illustration -->
|
|
<g transform="translate(272, 158)">
|
|
<rect x="0" y="0" width="26" height="18" rx="2" fill="#f39c12" fill-opacity="0.2" stroke="#f39c12" stroke-width="0.8"/>
|
|
<text x="13" y="12" text-anchor="middle" font-size="6" fill="#f39c12">0.7|0.3</text>
|
|
<rect x="30" y="0" width="26" height="18" rx="2" fill="#f39c12" fill-opacity="0.2" stroke="#f39c12" stroke-width="0.8"/>
|
|
<text x="43" y="12" text-anchor="middle" font-size="6" fill="#f39c12">0.9|0.1</text>
|
|
<rect x="60" y="0" width="26" height="18" rx="2" fill="#f39c12" fill-opacity="0.2" stroke="#f39c12" stroke-width="0.8"/>
|
|
<text x="73" y="12" text-anchor="middle" font-size="6" fill="#f39c12">0.5|0.5</text>
|
|
</g>
|
|
<text x="325" y="196" text-anchor="middle" font-size="9" fill="#666">Gumbel-Softmax / dVAE</text>
|
|
|
|
<!-- Middle examples -->
|
|
<rect x="255" y="216" width="140" height="44" rx="6" fill="#f39c12" fill-opacity="0.05" stroke="#f39c12" stroke-width="0.8" stroke-dasharray="3,2"/>
|
|
<text x="325" y="232" text-anchor="middle" font-size="8" fill="#f39c12">DALL-E (training),</text>
|
|
<text x="325" y="244" text-anchor="middle" font-size="8" fill="#f39c12">Maskbit</text>
|
|
|
|
<!-- Right: Continuous Tokens box -->
|
|
<rect x="420" y="128" width="200" height="80" rx="8" fill="#3498db" fill-opacity="0.1" stroke="#3498db" stroke-width="1.5"/>
|
|
<text x="520" y="150" text-anchor="middle" font-size="12" font-weight="bold" fill="#3498db">Use Continuous Tokens</text>
|
|
<!-- Continuous latent illustration -->
|
|
<g transform="translate(454, 158)">
|
|
<rect x="0" y="0" width="26" height="18" rx="2" fill="#3498db" fill-opacity="0.2" stroke="#3498db" stroke-width="0.8"/>
|
|
<text x="13" y="12" text-anchor="middle" font-size="6" fill="#3498db">0.73</text>
|
|
<rect x="30" y="0" width="26" height="18" rx="2" fill="#3498db" fill-opacity="0.2" stroke="#3498db" stroke-width="0.8"/>
|
|
<text x="43" y="12" text-anchor="middle" font-size="6" fill="#3498db">-0.21</text>
|
|
<rect x="60" y="0" width="26" height="18" rx="2" fill="#3498db" fill-opacity="0.2" stroke="#3498db" stroke-width="0.8"/>
|
|
<text x="73" y="12" text-anchor="middle" font-size="6" fill="#3498db">1.45</text>
|
|
<rect x="90" y="0" width="26" height="18" rx="2" fill="#3498db" fill-opacity="0.2" stroke="#3498db" stroke-width="0.8"/>
|
|
<text x="103" y="12" text-anchor="middle" font-size="6" fill="#3498db">-0.58</text>
|
|
</g>
|
|
<text x="520" y="196" text-anchor="middle" font-size="9" fill="#666">VAE latents (KL-regularised)</text>
|
|
|
|
<!-- Right examples -->
|
|
<rect x="430" y="216" width="180" height="44" rx="6" fill="#3498db" fill-opacity="0.05" stroke="#3498db" stroke-width="0.8" stroke-dasharray="3,2"/>
|
|
<text x="520" y="232" text-anchor="middle" font-size="8" fill="#3498db">Stable Diffusion, DALL-E 3,</text>
|
|
<text x="520" y="244" text-anchor="middle" font-size="8" fill="#3498db">Sora, Imagen Video</text>
|
|
|
|
<!-- Bottom note -->
|
|
<text x="325" y="284" text-anchor="middle" font-size="9" fill="#999">The generation method determines whether discrete or continuous tokenisation is most appropriate</text>
|
|
</svg> |