{ "type": "excalidraw", "version": 2, "source": "https://excalidraw.com", "elements": [ { "id": "text-0001", "type": "text", "x": 80, "y": 40, "width": 1540, "height": 44, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19000, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "常用 KV Cache 量化方法对比与选型", "fontSize": 30, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "常用 KV Cache 量化方法对比与选型", "autoResize": false, "lineHeight": 1.25 }, { "id": "text-0002", "type": "text", "x": 80, "y": 90, "width": 1540, "height": 32, "angle": 0, "strokeColor": "#64748b", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19097, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "按工程默认、研究低比特、极限长上下文三个层次理解", "fontSize": 17, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "按工程默认、研究低比特、极限长上下文三个层次理解", "autoResize": false, "lineHeight": 1.25 }, { "id": "arrow-0003", "type": "arrow", "x": 90, "y": 132, "width": 1520, "height": 0, "angle": 0, "strokeColor": "#94a3b8", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "dashed", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19194, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "points": [ [ 0, 0 ], [ 1520, 0 ] ], "lastCommittedPoint": null, "startBinding": null, "endBinding": null, "startArrowhead": null, "endArrowhead": null }, { "id": "rectangle-0004", "type": "rectangle", "x": 65, "y": 180, "width": 210, "height": 58, "angle": 0, "strokeColor": "#1d4ed8", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 19291, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0005", "type": "text", "x": 77, "y": 190, "width": 186, "height": 38, "angle": 0, "strokeColor": "#1e3a8a", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19388, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "方法", "fontSize": 19, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "方法", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0006", "type": "rectangle", "x": 275, "y": 180, "width": 410, "height": 58, "angle": 0, "strokeColor": "#1d4ed8", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 19485, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0007", "type": "text", "x": 287, "y": 190, "width": 386, "height": 38, "angle": 0, "strokeColor": "#1e3a8a", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19582, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "核心做法", "fontSize": 19, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "核心做法", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0008", "type": "rectangle", "x": 685, "y": 180, "width": 190, "height": 58, "angle": 0, "strokeColor": "#1d4ed8", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 19679, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0009", "type": "text", "x": 697, "y": 190, "width": 166, "height": 38, "angle": 0, "strokeColor": "#1e3a8a", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19776, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "常见比特", "fontSize": 19, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "常见比特", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0010", "type": "rectangle", "x": 875, "y": 180, "width": 360, "height": 58, "angle": 0, "strokeColor": "#1d4ed8", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 19873, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0011", "type": "text", "x": 887, "y": 190, "width": 336, "height": 38, "angle": 0, "strokeColor": "#1e3a8a", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 19970, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "优势", "fontSize": 19, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "优势", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0012", "type": "rectangle", "x": 1235, "y": 180, "width": 410, "height": 58, "angle": 0, "strokeColor": "#1d4ed8", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 20067, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0013", "type": "text", "x": 1247, "y": 190, "width": 386, "height": 38, "angle": 0, "strokeColor": "#1e3a8a", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 20164, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "风险和适用场景", "fontSize": 19, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "风险和适用场景", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0014", "type": "rectangle", "x": 65, "y": 238, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 20261, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0015", "type": "text", "x": 77, "y": 248, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 20358, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "FP8 KV\nvLLM / TensorRT-LLM", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "FP8 KV\nvLLM / TensorRT-LLM", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0016", "type": "rectangle", "x": 275, "y": 238, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 20455, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0017", "type": "text", "x": 287, "y": 248, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 20552, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "K/V 用 FP8 存储；scale 可 per-tensor 或 per-head；部分后端在 FP8 域算 attention", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "K/V 用 FP8 存储；scale 可 per-tensor 或 per-head；部分后端在 FP8 域算 attention", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0018", "type": "rectangle", "x": 685, "y": 238, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 20649, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0019", "type": "text", "x": 697, "y": 248, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 20746, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "8 bit", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "8 bit", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0020", "type": "rectangle", "x": 875, "y": 238, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 20843, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0021", "type": "text", "x": 887, "y": 248, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 20940, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "部署最成熟；通常 2x cache 容量；延迟和吞吐更可预测", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "部署最成熟；通常 2x cache 容量；延迟和吞吐更可预测", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0022", "type": "rectangle", "x": 1235, "y": 238, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 21037, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0023", "type": "text", "x": 1247, "y": 248, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 21134, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "压缩率有限；需要校准 scale 才更稳；先作为生产默认基线", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "压缩率有限；需要校准 scale 才更稳；先作为生产默认基线", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0024", "type": "rectangle", "x": 65, "y": 358, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 21231, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0025", "type": "text", "x": 77, "y": 368, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 21328, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "HF QuantizedCache\nQuanto / HQQ", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "HF QuantizedCache\nQuanto / HQQ", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0026", "type": "rectangle", "x": 275, "y": 358, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 21425, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0027", "type": "text", "x": 287, "y": 368, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 21522, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Transformers generate 中量化 cache；可配 backend、axis、residual length", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Transformers generate 中量化 cache；可配 backend、axis、residual length", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0028", "type": "rectangle", "x": 685, "y": 358, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 21619, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0029", "type": "text", "x": 697, "y": 368, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 21716, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "int2 / int4 / int8", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "int2 / int4 / int8", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0030", "type": "rectangle", "x": 875, "y": 358, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 21813, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0031", "type": "text", "x": 887, "y": 368, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 21910, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "上手快；适合单机实验和 OOM 缓解", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "上手快；适合单机实验和 OOM 缓解", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0032", "type": "rectangle", "x": 1235, "y": 358, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22007, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0033", "type": "text", "x": 1247, "y": 368, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 22104, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "短上下文可能变慢；backend 轴设置不同；需要留高精度残差窗口", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "短上下文可能变慢；backend 轴设置不同；需要留高精度残差窗口", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0034", "type": "rectangle", "x": 65, "y": 478, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22201, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0035", "type": "text", "x": 77, "y": 488, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 22298, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "KIVI", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "KIVI", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0036", "type": "rectangle", "x": 275, "y": 478, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22395, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0037", "type": "text", "x": 287, "y": 488, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 22492, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "K per-channel，V per-token；利用 K 的 channel outlier 分布差异", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "K per-channel，V per-token；利用 K 的 channel outlier 分布差异", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0038", "type": "rectangle", "x": 685, "y": 478, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22589, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0039", "type": "text", "x": 697, "y": 488, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 22686, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "2 bit", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "2 bit", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0040", "type": "rectangle", "x": 875, "y": 478, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22783, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0041", "type": "text", "x": 887, "y": 488, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 22880, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "调参少；是低比特 KV 量化的经典基线", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "调参少；是低比特 KV 量化的经典基线", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0042", "type": "rectangle", "x": 1235, "y": 478, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 22977, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0043", "type": "text", "x": 1247, "y": 488, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 23074, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "工程实现依赖 kernel；长推理任务仍需业务验证", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "工程实现依赖 kernel；长推理任务仍需业务验证", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0044", "type": "rectangle", "x": 65, "y": 598, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 23171, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0045", "type": "text", "x": 77, "y": 608, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 23268, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "KVQuant", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "KVQuant", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0046", "type": "rectangle", "x": 275, "y": 598, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 23365, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0047", "type": "text", "x": 287, "y": 608, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 23462, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Pre-RoPE K、non-uniform dtype、dense-sparse outlier、Q-Norm", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Pre-RoPE K、non-uniform dtype、dense-sparse outlier、Q-Norm", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0048", "type": "rectangle", "x": 685, "y": 598, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 23559, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0049", "type": "text", "x": 697, "y": 608, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 23656, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "2 / 3 bit", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "2 / 3 bit", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0050", "type": "rectangle", "x": 875, "y": 598, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 23753, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0051", "type": "text", "x": 887, "y": 608, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 23850, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "针对超长上下文；论文展示 1M 到 10M context 潜力", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "针对超长上下文；论文展示 1M 到 10M context 潜力", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0052", "type": "rectangle", "x": 1235, "y": 598, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 23947, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0053", "type": "text", "x": 1247, "y": 608, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 24044, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "实现复杂；需要自定义 CUDA；更像研究和专用系统路线", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "实现复杂；需要自定义 CUDA；更像研究和专用系统路线", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0054", "type": "rectangle", "x": 65, "y": 718, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 24141, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0055", "type": "text", "x": 77, "y": 728, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 24238, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "GEAR / SKVQ / WKVQuant", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "GEAR / SKVQ / WKVQuant", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0056", "type": "rectangle", "x": 275, "y": 718, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 24335, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0057", "type": "text", "x": 287, "y": 728, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 24432, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "量化叠加低秩误差、稀疏 outlier、滑动窗口或权重+KV 联合 PTQ", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "量化叠加低秩误差、稀疏 outlier、滑动窗口或权重+KV 联合 PTQ", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0058", "type": "rectangle", "x": 685, "y": 718, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 24529, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0059", "type": "text", "x": 697, "y": 728, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 24626, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "约 1.5 到 4 bit", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "约 1.5 到 4 bit", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0060", "type": "rectangle", "x": 875, "y": 718, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 24723, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0061", "type": "text", "x": 887, "y": 728, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 24820, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "针对更高压缩率或联合优化，提供很多设计思路", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "针对更高压缩率或联合优化，提供很多设计思路", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0062", "type": "rectangle", "x": 1235, "y": 718, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 24917, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0063", "type": "text", "x": 1247, "y": 728, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25014, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "方法各有额外假设；迁移到 vLLM/sglang 需要 kernel 和调度配合", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "方法各有额外假设；迁移到 vLLM/sglang 需要 kernel 和调度配合", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0064", "type": "rectangle", "x": 65, "y": 838, "width": 210, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 25111, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0065", "type": "text", "x": 77, "y": 848, "width": 186, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25208, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "TurboQuant", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "TurboQuant", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0066", "type": "rectangle", "x": 275, "y": 838, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 25305, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0067", "type": "text", "x": 287, "y": 848, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25402, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "旋转后逐坐标标量量化；论文加 QJL 残差修正；vLLM preset 加 norm correction", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "旋转后逐坐标标量量化；论文加 QJL 残差修正；vLLM preset 加 norm correction", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0068", "type": "rectangle", "x": 685, "y": 838, "width": 190, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 25499, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0069", "type": "text", "x": 697, "y": 848, "width": 166, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25596, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "3 / 4 bit\nk8v4", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "3 / 4 bit\nk8v4", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0070", "type": "rectangle", "x": 875, "y": 838, "width": 360, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 25693, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0071", "type": "text", "x": 887, "y": 848, "width": 336, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25790, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "在显存紧张时扩大可服务 token 容量；算法解释性强", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "在显存紧张时扩大可服务 token 容量；算法解释性强", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0072", "type": "rectangle", "x": 1235, "y": 838, "width": 410, "height": 120, "angle": 0, "strokeColor": "#cbd5e1", "backgroundColor": "#f8fafc", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 25887, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0073", "type": "text", "x": 1247, "y": 848, "width": 386, "height": 100, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 25984, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "当前 vLLM 研究显示 FP8 仍是默认首选；4bit_nc 可试，3bit 需要谨慎", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "当前 vLLM 研究显示 FP8 仍是默认首选；4bit_nc 可试，3bit 需要谨慎", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0074", "type": "rectangle", "x": 160, "y": 980, "width": 1380, "height": 85, "angle": 0, "strokeColor": "#d97706", "backgroundColor": "#fef3c7", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 26081, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0075", "type": "text", "x": 172, "y": 990, "width": 1356, "height": 65, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 26178, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "实战决策树：先确认是不是 KV cache 限制吞吐或上下文；不是就别急着压 KV。是的话先测 FP8；仍 OOM 或排队严重，再比较 4bit / KIVI / TurboQuant / KVQuant。", "fontSize": 20, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "实战决策树：先确认是不是 KV cache 限制吞吐或上下文；不是就别急着压 KV。是的话先测 FP8；仍 OOM 或排队严重，再比较 4bit / KIVI / TurboQuant / KVQuant。", "autoResize": false, "lineHeight": 1.25 }, { "id": "rectangle-0076", "type": "rectangle", "x": 260, "y": 1090, "width": 1180, "height": 64, "angle": 0, "strokeColor": "#0f766e", "backgroundColor": "#99f6e4", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 26275, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "text-0077", "type": "text", "x": 272, "y": 1100, "width": 1156, "height": 44, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 26372, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "评测必须覆盖：VRAM 峰值、TTFT、TPOT、throughput、cache hit、困惑度、长上下文检索、数学/代码推理、目标业务样本。", "fontSize": 18, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "评测必须覆盖：VRAM 峰值、TTFT、TPOT、throughput、cache hit、困惑度、长上下文检索、数学/代码推理、目标业务样本。", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-bg", "type": "rectangle", "x": 80, "y": 1234.0, "width": 1565.0, "height": 300, "angle": 0, "strokeColor": "#64748b", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 70, "groupIds": [], "seed": 93001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-title", "type": "text", "x": 108, "y": 1256.0, "width": 1509.0, "height": 38, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "练习题与面试题补充：KV cache 量化", "fontSize": 24, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "练习题与面试题补充：KV cache 量化", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-practice-box", "type": "rectangle", "x": 108, "y": 1312.0, "width": 740.5, "height": 190, "angle": 0, "strokeColor": "#2563eb", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 93002, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-practice-text", "type": "text", "x": 130, "y": 1332.0, "width": 717.5, "height": 150, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "练习题\n1. 计算不同 batch/context/kv_heads 下的 KV cache 显存\n2. 对比 FP8、INT4、2bit 在显存、质量、kernel 复杂度上的取舍\n3. 设计一套长上下文 needle 评测协议验证量化退化", "fontSize": 17, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "练习题\n1. 计算不同 batch/context/kv_heads 下的 KV cache 显存\n2. 对比 FP8、INT4、2bit 在显存、质量、kernel 复杂度上的取舍\n3. 设计一套长上下文 needle 评测协议验证量化退化", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-interview-box", "type": "rectangle", "x": 876.5, "y": 1312.0, "width": 740.5, "height": 190, "angle": 0, "strokeColor": "#7c3aed", "backgroundColor": "#ddd6fe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 93003, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-02-kvcache-quant-methods-comparison-interview-text", "type": "text", "x": 898.5, "y": 1332.0, "width": 717.5, "height": 150, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "面试题\n1. 为什么 K 和 V 的量化敏感性不同\n2. FP8 为什么常是工程默认首选而不是最低 bit\n3. TurboQuant 的旋转和 inner-product bias 目标解决什么问题", "fontSize": 17, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "面试题\n1. 为什么 K 和 V 的量化敏感性不同\n2. FP8 为什么常是工程默认首选而不是最低 bit\n3. TurboQuant 的旋转和 inner-product bias 目标解决什么问题", "autoResize": false, "lineHeight": 1.25 } ], "appState": { "gridSize": null, "viewBackgroundColor": "#ffffff" }, "files": {} }