From 06b87802d187e6ce03b055d8a1c258dc2a4c1811 Mon Sep 17 00:00:00 2001 From: huangzhuang_3rd Date: Mon, 1 Jun 2026 18:30:59 +0800 Subject: [PATCH] =?UTF-8?q?v1.8.0:=20=E6=A8=A1=E5=9D=97=E5=8C=96=E9=87=8D?= =?UTF-8?q?=E6=9E=84=20+=20=E5=90=8E=E5=8F=B0=E7=99=BB=E5=BD=95=E8=AE=A4?= =?UTF-8?q?=E8=AF=81?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- data/knowledge.json | 66 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 66 insertions(+) create mode 100644 data/knowledge.json diff --git a/data/knowledge.json b/data/knowledge.json new file mode 100644 index 0000000..882531d --- /dev/null +++ b/data/knowledge.json @@ -0,0 +1,66 @@ +[ + { + "id": "k001", + "title": "什么是参数量?", + "category": "ai-models", + "icon": "ri-calculator-line", + "content": "参数量(Parameters)是衡量大模型规模的指标,表示模型中权重参数的数量。例如 GPT-3 有 175B 参数,即约1750亿个参数。", + "detail": "参数量决定了模型的容量和表达能力。一般来说,参数量越大,模型能力越强,但也需要更多计算资源。\n\n常见规模分类:\n- 小模型:<1B (适合边缘设备)\n- 中模型:1B-10B (消费级GPU可运行)\n- 大模型:10B-100B (需要多GPU)\n- 超大模型:>100B (需要数据中心)", + "order": 1 + }, + { + "id": "k002", + "title": "什么是上下文长度?", + "category": "ai-models", + "icon": "ri-text-wrap", + "content": "上下文长度(Context Length)是模型能处理的输入文本最大长度。更长的上下文意味着模型可以理解更长的文档或对话历史。", + "detail": "常见长度:\n- 4K:传统长度,适合简单对话\n- 32K:中等长度,适合长文档\n- 128K:超长上下文,如GPT-4 Turbo\n- 200K:Claude 3的极限长度", + "order": 2 + }, + { + "id": "k003", + "title": "什么是量化?", + "category": "ai-models", + "icon": "ri-scales-3-line", + "content": "量化(Quantization)是将模型参数从高精度转换为低精度,减少显存占用和计算量。如FP16→INT8→INT4,精度损失可控,资源节省显著。", + "detail": "量化效果:\n- FP32→FP16: 显存减半,精度基本不变\n- FP16→INT8: 显存再减半,精度略降\n- INT8→INT4: 显存再减半,需特殊技术\n\n推荐工具:llama.cpp、GPTQ、AWQ等", + "order": 3 + }, + { + "id": "k004", + "title": "什么是MMLU?", + "category": "ai-models", + "icon": "ri-bar-chart-box-line", + "content": "MMLU(Massive Multitask Language Understanding)是评估大模型综合能力的标准测试集,覆盖57个学科领域。", + "detail": "分数参考:\n- 60-70%:入门级,如GPT-3\n- 70-80%:中等水平,如Llama 2 70B\n- 80-90%:优秀水平,如GPT-4、Claude 3", + "order": 4, + "visible": false + }, + { + "id": "k005", + "title": "如何计算显存需求?", + "category": "gpus", + "icon": "ri-memory-line", + "content": "模型显存需求 ≈ 参数量 × 每参数字节数 × 1.3(含KV Cache开销)", + "detail": "计算公式:\n- FP32: 参数量 × 4字节 × 1.3\n- FP16: 参数量 × 2字节 × 1.3\n- INT8: 参数量 × 1字节 × 1.3\n- INT4: 参数量 × 0.5字节 × 1.3\n\n例如:7B模型FP16加载需要约 7 × 2 × 1.3 ≈ 18GB显存", + "order": 1 + }, + { + "id": "k006", + "title": "GPU架构演进", + "category": "gpus", + "icon": "ri-history-line", + "content": "NVIDIA GPU架构从Fermi到Hopper,每一代都有显著提升。了解架构有助于选择合适的GPU。", + "detail": "主要架构:\n- Volta (2017): V100, 引入Tensor Core\n- Turing (2018): RTX 20系列, RT Core\n- Ampere (2020): A100, RTX 30系列\n- Hopper (2022): H100, FP8支持\n- Ada Lovelace (2022): RTX 40系列, L40S", + "order": 2 + }, + { + "id": "k007", + "title": "CPU核心数选择", + "category": "cpus", + "icon": "ri-database-2-line", + "content": "CPU核心数的选择取决于应用场景。更多核心适合并行任务,但单核性能也很重要。", + "detail": "场景推荐:\n- 办公/日常:4-6核足够\n- 开发/编译:8-16核\n- 服务器/虚拟化:16-64核\n- 高性能计算:64核以上\n\n注意:AI训练主要依赖GPU,CPU主要用于数据预处理", + "order": 1 + } +] \ No newline at end of file