| 1 | <?xml version='1.0' encoding='UTF-8'?> |
| 2 | <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1280 720" width="1280" height="720"> |
| 3 | <defs> |
| 4 | <filter id="cardShadow" x="-10%" y="-10%" width="120%" height="120%"> |
| 5 | <feGaussianBlur in="SourceAlpha" stdDeviation="4" /> |
| 6 | <feOffset dx="0" dy="3" /> |
| 7 | <feFlood flood-color="#1A365D" flood-opacity="0.07" /> |
| 8 | <feComposite in2="SourceAlpha" operator="in" /> |
| 9 | <feMerge><feMergeNode /><feMergeNode in="SourceGraphic" /></feMerge> |
| 10 | </filter> |
| 11 | </defs> |
| 12 | |
| 13 | <g id="bg"> |
| 14 | <rect width="1280" height="720" fill="#FFFFFF" /> |
| 15 | </g> |
| 16 | |
| 17 | <g id="header"> |
| 18 | <rect x="60" y="56" width="6" height="40" fill="#3182CE" /> |
| 19 | <text x="84" y="88" font-family="Georgia, "Microsoft YaHei", serif" font-size="36" font-weight="700" fill="#1A365D">A simpler model, better results, less training</text> |
| 20 | <text x="84" y="118" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#4A5568" letter-spacing="1.5">WHY THIS PAPER MATTERS · 2017 SOTA ON WMT'14 TRANSLATION</text> |
| 21 | </g> |
| 22 | |
| 23 | <g id="kpi-bleu-ende"> |
| 24 | <rect x="60" y="160" width="285" height="240" rx="12" fill="#FFFFFF" stroke="#E2E8F0" stroke-width="1.5" filter="url(#cardShadow)" /> |
| 25 | <rect x="60" y="160" width="6" height="240" rx="3" fill="#1A365D" /> |
| 26 | <use data-icon="tabler-outline/trophy" x="80" y="184" width="32" height="32" fill="#1A365D" /> |
| 27 | <text x="80" y="246" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#4A5568" letter-spacing="1">EN → DE</text> |
| 28 | <text x="80" y="270" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="14" fill="#4A5568">BLEU on WMT'14</text> |
| 29 | <text x="80" y="336" font-family="Georgia, "Microsoft YaHei", serif" font-size="64" font-weight="700" fill="#1A365D">28.4</text> |
| 30 | <use data-icon="tabler-outline/arrow-up-right" x="80" y="354" width="20" height="20" fill="#2F855A" /> |
| 31 | <text x="106" y="370" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="16" font-weight="700" fill="#2F855A">+2.0</text> |
| 32 | <text x="150" y="370" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" fill="#4A5568">over prior ensembles</text> |
| 33 | </g> |
| 34 | |
| 35 | <g id="kpi-bleu-enfr"> |
| 36 | <rect x="365" y="160" width="285" height="240" rx="12" fill="#FFFFFF" stroke="#E2E8F0" stroke-width="1.5" filter="url(#cardShadow)" /> |
| 37 | <rect x="365" y="160" width="6" height="240" rx="3" fill="#3182CE" /> |
| 38 | <use data-icon="tabler-outline/trophy" x="385" y="184" width="32" height="32" fill="#3182CE" /> |
| 39 | <text x="385" y="246" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#4A5568" letter-spacing="1">EN → FR</text> |
| 40 | <text x="385" y="270" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="14" fill="#4A5568">BLEU on WMT'14</text> |
| 41 | <text x="385" y="336" font-family="Georgia, "Microsoft YaHei", serif" font-size="64" font-weight="700" fill="#3182CE">41.8</text> |
| 42 | <use data-icon="tabler-outline/star" x="385" y="354" width="20" height="20" fill="#2F855A" /> |
| 43 | <text x="411" y="370" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#2F855A">new single-model SOTA</text> |
| 44 | </g> |
| 45 | |
| 46 | <g id="kpi-training"> |
| 47 | <rect x="670" y="160" width="285" height="240" rx="12" fill="#FFFFFF" stroke="#E2E8F0" stroke-width="1.5" filter="url(#cardShadow)" /> |
| 48 | <rect x="670" y="160" width="6" height="240" rx="3" fill="#1A365D" /> |
| 49 | <use data-icon="tabler-outline/bolt" x="690" y="184" width="32" height="32" fill="#1A365D" /> |
| 50 | <text x="690" y="246" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#4A5568" letter-spacing="1">TRAINING TIME</text> |
| 51 | <text x="690" y="270" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="14" fill="#4A5568">Transformer (big)</text> |
| 52 | <text x="690" y="336" font-family="Georgia, "Microsoft YaHei", serif" font-size="48" font-weight="700" fill="#1A365D">3.5 days</text> |
| 53 | <use data-icon="tabler-outline/cpu" x="690" y="354" width="20" height="20" fill="#4A5568" /> |
| 54 | <text x="716" y="370" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" fill="#4A5568">on 8× P100 GPUs</text> |
| 55 | </g> |
| 56 | |
| 57 | <g id="kpi-no-rnn-cnn"> |
| 58 | <rect x="975" y="160" width="245" height="240" rx="12" fill="#FFFFFF" stroke="#E2E8F0" stroke-width="1.5" filter="url(#cardShadow)" /> |
| 59 | <rect x="975" y="160" width="6" height="240" rx="3" fill="#3182CE" /> |
| 60 | <use data-icon="tabler-outline/eye" x="995" y="184" width="32" height="32" fill="#3182CE" /> |
| 61 | <text x="995" y="246" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#4A5568" letter-spacing="1">ARCHITECTURE</text> |
| 62 | <text x="995" y="270" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="14" fill="#4A5568">building blocks</text> |
| 63 | <text x="995" y="316" font-family="Georgia, "Microsoft YaHei", serif" font-size="40" font-weight="700" fill="#1A365D">Attention</text> |
| 64 | <text x="995" y="346" font-family="Georgia, "Microsoft YaHei", serif" font-size="22" font-style="italic" fill="#4A5568">only</text> |
| 65 | <text x="995" y="376" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" fill="#4A5568">no recurrence · no convolution</text> |
| 66 | </g> |
| 67 | |
| 68 | <g id="takeaway-band"> |
| 69 | <rect x="60" y="450" width="1160" height="160" rx="12" fill="#F5F7FA" stroke="#E2E8F0" stroke-width="1" /> |
| 70 | <rect x="60" y="450" width="6" height="160" rx="3" fill="#1A365D" /> |
| 71 | <text x="100" y="490" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="13" font-weight="700" fill="#3182CE" letter-spacing="1.5">THE HEADLINE</text> |
| 72 | <text x="100" y="520" font-family="Georgia, "Microsoft YaHei", serif" font-size="22" font-weight="700" fill="#1A365D"> |
| 73 | The Transformer reaches <tspan fill="#3182CE">new translation SOTA</tspan> while training |
| 74 | </text> |
| 75 | <text x="100" y="548" font-family="Georgia, "Microsoft YaHei", serif" font-size="22" font-weight="700" fill="#1A365D"> |
| 76 | in <tspan fill="#3182CE">a fraction of the time</tspan> of strong baselines. |
| 77 | </text> |
| 78 | <text x="100" y="580" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="16" fill="#4A5568"> |
| 79 | Quality up · cost down · parallelization unlocked. The whole rest of the paper explains |
| 80 | </text> |
| 81 | <text x="100" y="602" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="16" fill="#4A5568"> |
| 82 | how attention alone makes this possible — and why nothing else is needed. |
| 83 | </text> |
| 84 | </g> |
| 85 | |
| 86 | <g id="footer"> |
| 87 | <text x="60" y="694" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="11" fill="#A0AEC0">Source: Vaswani et al. 2017, Table 2 (BLEU) and §5.2 (training cost)</text> |
| 88 | <text x="1220" y="694" text-anchor="end" font-family="Arial, "Microsoft YaHei", sans-serif" font-size="11" fill="#A0AEC0">02</text> |
| 89 | </g> |
| 90 | </svg> |
| 91 |