liqiang
b119135836
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
79 lines
9.1 KiB
XML
79 lines
9.1 KiB
XML
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 980 620" width="980" height="620" style="background:#ffffff">
|
||
<defs><marker id="arrow" markerWidth="12" markerHeight="8" refX="11" refY="4" orient="auto"><polygon points="0 0, 12 4, 0 8" fill="#333333"/></marker></defs>
|
||
<text x="490" y="25" font-family="Arial, sans-serif" font-size="19" fill="#333333" text-anchor="middle" font-weight="bold">同一 SWE-bench 任务：16 次 rollout 如何产生一次更新</text>
|
||
|
||
<rect x="30" y="65" width="190" height="120" rx="7" fill="#f0f0f0" stroke="#333333" stroke-width="2"/>
|
||
<text x="125" y="88" font-family="Arial, sans-serif" font-size="16" fill="#333333" text-anchor="middle" font-weight="bold">同一任务 + 初始代码</text>
|
||
<text x="45" y="116" font-family="'Courier New', monospace" font-size="13" fill="#333333">parser.py</text>
|
||
<text x="45" y="137" font-family="'Courier New', monospace" font-size="13" fill="#333333">空输入触发 IndexError</text>
|
||
<text x="45" y="158" font-family="'Courier New', monospace" font-size="13" fill="#333333">不能修改测试文件</text>
|
||
|
||
<line x1="222" y1="125" x2="260" y2="125" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<rect x="262" y="75" width="170" height="100" rx="7" fill="#d9d9d9" stroke="#333333" stroke-width="2"/>
|
||
<text x="347" y="101" font-family="Arial, sans-serif" font-size="17" fill="#333333" text-anchor="middle" font-weight="bold">策略模型</text>
|
||
<text x="347" y="128" font-family="Arial, sans-serif" font-size="14" fill="#333333" text-anchor="middle">独立采样 16 条</text>
|
||
<text x="347" y="151" font-family="Arial, sans-serif" font-size="13" fill="#666666" text-anchor="middle">每条都是一次完整执行</text>
|
||
|
||
<line x1="434" y1="125" x2="468" y2="125" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<rect x="460" y="42" width="500" height="255" rx="7" fill="#fafafa" stroke="#333333" stroke-width="2"/>
|
||
<text x="710" y="64" font-family="Arial, sans-serif" font-size="16" fill="#333333" text-anchor="middle" font-weight="bold">16 个相同沙箱中的 rollout（采样不同，初始状态相同）</text>
|
||
<rect x="480" y="82" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="532.5" y="98" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#01 × 失败</text>
|
||
<rect x="595" y="82" width="105" height="30" rx="5" fill="#666666" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="647.5" y="98" font-family="Arial, sans-serif" font-size="12" fill="#ffffff" text-anchor="middle" dominant-baseline="central">#02 ✓ 通过</text>
|
||
<rect x="710" y="82" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="762.5" y="98" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#03 × 失败</text>
|
||
<rect x="825" y="82" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="877.5" y="98" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#04 × 失败</text>
|
||
<rect x="480" y="124" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="532.5" y="140" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#05 × 失败</text>
|
||
<rect x="595" y="124" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="647.5" y="140" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#06 × 失败</text>
|
||
<rect x="710" y="124" width="105" height="30" rx="5" fill="#666666" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="762.5" y="140" font-family="Arial, sans-serif" font-size="12" fill="#ffffff" text-anchor="middle" dominant-baseline="central">#07 ✓ 通过</text>
|
||
<rect x="825" y="124" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="877.5" y="140" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#08 × 失败</text>
|
||
<rect x="480" y="166" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="532.5" y="182" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#09 × 失败</text>
|
||
<rect x="595" y="166" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="647.5" y="182" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#10 × 失败</text>
|
||
<rect x="710" y="166" width="105" height="30" rx="5" fill="#666666" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="762.5" y="182" font-family="Arial, sans-serif" font-size="12" fill="#ffffff" text-anchor="middle" dominant-baseline="central">#11 ✓ 通过</text>
|
||
<rect x="825" y="166" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="877.5" y="182" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#12 × 失败</text>
|
||
<rect x="480" y="208" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="532.5" y="224" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#13 × 失败</text>
|
||
<rect x="595" y="208" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="647.5" y="224" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#14 × 失败</text>
|
||
<rect x="710" y="208" width="105" height="30" rx="5" fill="#eeeeee" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="762.5" y="224" font-family="Arial, sans-serif" font-size="12" fill="#333333" text-anchor="middle" dominant-baseline="central">#15 × 失败</text>
|
||
<rect x="825" y="208" width="105" height="30" rx="5" fill="#666666" stroke="#333333" stroke-width="1.5"/>
|
||
<text x="877.5" y="224" font-family="Arial, sans-serif" font-size="12" fill="#ffffff" text-anchor="middle" dominant-baseline="central">#16 ✓ 通过</text>
|
||
<text x="710" y="267" font-family="Arial, sans-serif" font-size="13" fill="#666666" text-anchor="middle">4 条通过，12 条失败</text>
|
||
|
||
<line x1="710" y1="299" x2="710" y2="330" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<rect x="240" y="332" width="720" height="72" rx="7" fill="#f0f0f0" stroke="#333333" stroke-width="2"/>
|
||
<text x="600" y="354" font-family="Arial, sans-serif" font-size="16" fill="#333333" text-anchor="middle" font-weight="bold">奖励计算：验证器运行隐藏测试并检查路径</text>
|
||
<text x="420" y="382" font-family="Arial, sans-serif" font-size="14" fill="#333333" text-anchor="middle">通过全部测试且未改测试 → 奖励 1</text>
|
||
<text x="780" y="382" font-family="Arial, sans-serif" font-size="14" fill="#333333" text-anchor="middle">测试失败或违规 → 奖励 0</text>
|
||
|
||
<line x1="600" y1="406" x2="600" y2="435" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<rect x="240" y="437" width="720" height="70" rx="7" fill="#fafafa" stroke="#333333" stroke-width="2"/>
|
||
<text x="600" y="459" font-family="Arial, sans-serif" font-size="16" fill="#333333" text-anchor="middle" font-weight="bold">组内比较 → 相对优势</text>
|
||
<text x="420" y="486" font-family="Arial, sans-serif" font-size="14" fill="#333333" text-anchor="middle">4 条成功高于平均 → 正优势</text>
|
||
<text x="780" y="486" font-family="Arial, sans-serif" font-size="14" fill="#333333" text-anchor="middle">12 条失败低于平均 → 负优势</text>
|
||
|
||
<line x1="238" y1="472" x2="205" y2="472" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<rect x="30" y="360" width="175" height="145" rx="7" fill="#d9d9d9" stroke="#333333" stroke-width="2"/>
|
||
<text x="117.5" y="385" font-family="Arial, sans-serif" font-size="16" fill="#333333" text-anchor="middle" font-weight="bold">策略更新</text>
|
||
<text x="117.5" y="414" font-family="Arial, sans-serif" font-size="13" fill="#333333" text-anchor="middle">Gradient Descent</text>
|
||
<text x="117.5" y="434" font-family="Arial, sans-serif" font-size="13" fill="#333333" text-anchor="middle">+ Optimizer</text>
|
||
<text x="117.5" y="464" font-family="Arial, sans-serif" font-size="13" fill="#333333" text-anchor="middle">成功路径 ↑</text>
|
||
<text x="117.5" y="484" font-family="Arial, sans-serif" font-size="13" fill="#333333" text-anchor="middle">失败路径 ↓</text>
|
||
|
||
<path d="M117 358 L117 225 L245 225 L245 125 L260 125" fill="none" stroke="#333333" stroke-width="2" marker-end="url(#arrow)"/>
|
||
<text x="128" y="215" font-family="Arial, sans-serif" font-size="13" fill="#666666">下一 step</text>
|
||
<text x="490" y="555" font-family="Arial, sans-serif" font-size="15" fill="#666666" text-anchor="middle">一个 step：rollout → 验证 → 比较 → 梯度下降更新</text>
|
||
<text x="490" y="583" font-family="Arial, sans-serif" font-size="14" fill="#666666" text-anchor="middle">粗略量级：rollout ≈ 2,000 s + update ≈ 600 s → step ≈ 2,600 s（约 43 分钟）</text>
|
||
</svg>
|
