<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>象洞学长的个人空间</title><description>象洞学长的个人网站，记录学习笔记、兴趣与生活。</description><link>https://example.com/</link><language>zh-CN</language><item><title>用卡卡字幕助手给口播加字幕，再强调关键词</title><link>https://example.com/blog/videocaptioner-auto-subtitles/</link><guid isPermaLink="true">https://example.com/blog/videocaptioner-auto-subtitles/</guid><description>记录 VideoCaptioner 的安装、自动识别与硬字幕导出流程，以及通过 ASS 局部样式将关键词放大、加粗的实践。</description><pubDate>Thu, 29 Oct 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>图像与视频工具</category><category>VideoCaptioner</category><category>自动字幕</category><category>口播视频</category><category>ASS</category><category>视频制作</category></item><item><title>笔记：知识视频提示词：人文纪录片与 3D 科普短片</title><link>https://example.com/notes/vibe-knowledge-video-prompts/</link><guid isPermaLink="true">https://example.com/notes/vibe-knowledge-video-prompts/</guid><description>收藏两套可复制的代码视频提示词：以锚词组织叙事的人文纪录片模板，以及包含事实核验、3D 机制演示和逐帧验收的科普短片模板。</description><pubDate>Mon, 26 Oct 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>AI 视频</category><category>提示词</category><category>知识科普</category><category>分镜</category><category>3D 动画</category></item><item><title>笔记：VAE：从自编码器到可采样的生成模型</title><link>https://example.com/notes/vae-from-ae-to-generative-model/</link><guid isPermaLink="true">https://example.com/notes/vae-from-ae-to-generative-model/</guid><description>从 AE、变分推断和重参数化理解 VAE，结合原始论文梳理 MNIST/Frey Face 实验、概率架构、ELBO 损失与训练策略。</description><pubDate>Sun, 25 Oct 2026 00:00:00 GMT</pubDate><category>读书笔记</category><category>VAE</category><category>自编码器</category><category>生成模型</category><category>变分推断</category><category>论文阅读</category></item><item><title>笔记：Qwen-VL：中英对照阅读笔记</title><link>https://example.com/notes/qwen-vl-bilingual/</link><guid isPermaLink="true">https://example.com/notes/qwen-vl-bilingual/</guid><description>左侧英文原文，右侧中文译文，逐段对照阅读 Qwen-VL 的模型架构、三阶段训练、实验与附录，保留原图并直链文内引用。</description><pubDate>Thu, 22 Oct 2026 00:00:00 GMT</pubDate><category>读书笔记</category><category>Qwen-VL</category><category>多模态</category><category>视觉语言模型</category><category>论文阅读</category></item><item><title>LoRA 与 QLoRA：用交互实验理解低秩更新与量化微调</title><link>https://example.com/blog/lora-qlora-interactive-guide/</link><guid isPermaLink="true">https://example.com/blog/lora-qlora-interactive-guide/</guid><description>拖动秩、逐步查看训练路径、对比显存账本，理解两个小矩阵如何完成微调，以及 QLoRA 怎样压缩冻结基座。</description><pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>微调与显存优化</category><category>大模型</category><category>模型微调</category><category>LoRA</category><category>QLoRA</category><category>显存优化</category></item><item><title>千问如何越跑越快：从 MHA、GQA 到混合注意力</title><link>https://example.com/blog/qwen-attention-inference-evolution/</link><guid isPermaLink="true">https://example.com/blog/qwen-attention-inference-evolution/</guid><description>沿着 Qwen 的公开结构变化，理解 KV cache 为什么拖慢生成，以及 GQA、Gated DeltaNet、MoE 与多 token 预测分别优化了什么。</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>模型基础与架构</category><category>深度学习</category><category>Qwen</category><category>注意力机制</category><category>GQA</category><category>KV Cache</category><category>模型加速</category></item><item><title>笔记：AI 知识视频制作：从 Remotion 入门到可复用的实践方法</title><link>https://example.com/notes/ai-knowledge-video-workflow/</link><guid isPermaLink="true">https://example.com/notes/ai-knowledge-video-workflow/</guid><description>记录一次约 100 元 Token 的 AI 知识视频实践，解释 Remotion、Claude/Codex/DeepSeek 的分工，并整理可复用的叙事、画面、音乐和验收方法。</description><pubDate>Sun, 11 Oct 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>AI 视频</category><category>提示词</category><category>知识科普</category><category>分镜</category><category>Remotion</category><category>实践复盘</category></item><item><title>笔记：Docker 入门实践：从网页工具箱到 GPU 视觉模型</title><link>https://example.com/notes/docker-getting-started/</link><guid isPermaLink="true">https://example.com/notes/docker-getting-started/</guid><description>Windows Docker 与 WSL 2 配置、Docker Desktop 界面图解、容器管理、端口和数据持久化，以及 IT-Tools、rembg GPU 部署与排错。</description><pubDate>Sun, 11 Oct 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Docker</category><category>WSL 2</category><category>GPU</category><category>计算机视觉</category><category>模型部署</category></item><item><title>容易混淆的 DINO：自监督 DINO/DINOv2、检测 DINO 与 Grounding DINO</title><link>https://example.com/blog/dino-names-explained/</link><guid isPermaLink="true">https://example.com/blog/dino-names-explained/</guid><description>从训练信号、架构与输出区分同名 DINO，理解 DINOv2 的自蒸馏、iBOT、骨干能力，以及它与 CLIP 和检测 DINO 的关系。</description><pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉理解</category><category>计算机视觉</category><category>DINOv2</category><category>DINO</category><category>DETR</category><category>自监督学习</category><category>目标检测</category><category>CLIP</category></item><item><title>图像生成架构演进：从 VAE、GAN 到 DDPM、Stable Diffusion、DiT 与 Flow Matching</title><link>https://example.com/blog/image-generation-evolution/</link><guid isPermaLink="true">https://example.com/blog/image-generation-evolution/</guid><description>分清生成目标、像素与潜空间、U-Net 与 Transformer，系统理解 GAN、扩散模型、LDM、Stable Diffusion、DiT 与流匹配生成架构，并看清流匹配当前在图像、视频、音频、3D 和机器人上的落地位置。</description><pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>图像生成</category><category>图像生成</category><category>GAN</category><category>DDPM</category><category>Latent Diffusion</category><category>Stable Diffusion</category><category>DiT</category><category>Flow Matching</category></item><item><title>DETR 目标检测方法：从 DETR、Deformable DETR 到 DINO 与 RT-DETR</title><link>https://example.com/blog/detr-evolution/</link><guid isPermaLink="true">https://example.com/blog/detr-evolution/</guid><description>从集合预测、稀疏注意力、动态锚框和去噪训练，到高精度 DINO 与实时 RT-DETR，梳理 DETR 系列的演进、指标边界与部署选择。</description><pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉理解</category><category>目标检测</category><category>DETR</category><category>Deformable DETR</category><category>DAB-DETR</category><category>DN-DETR</category><category>DINO</category><category>RT-DETR</category></item><item><title>图片格式怎么选：JPEG、PNG、SVG、GIF、WebP、RAW、TIFF 与 ICO</title><link>https://example.com/blog/image-formats-guide/</link><guid isPermaLink="true">https://example.com/blog/image-formats-guide/</guid><description>从有损与无损、透明与动画、像素与矢量出发，理解八类常见图片格式的用途，并掌握网页配图、格式转换和压缩的实际方法。</description><pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>图像与视频工具</category><category>图片格式</category><category>JPEG</category><category>PNG</category><category>SVG</category><category>WebP</category><category>图像压缩</category><category>网站开发</category></item><item><title>一个网址从输入到打开，到底发生了什么：网页加载与文件下载全链路</title><link>https://example.com/blog/url-to-page-and-download/</link><guid isPermaLink="true">https://example.com/blog/url-to-page-and-download/</guid><description>沿着 URL、DNS、TCP/QUIC、TLS、HTTP 与浏览器渲染梳理网页加载，并解释 CDN、流式下载、Range 断点续传和分层故障排查。</description><pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>网络基础</category><category>网络基础</category><category>DNS</category><category>TCP</category><category>TLS</category><category>HTTP</category><category>浏览器</category><category>CDN</category><category>文件下载</category></item><item><title>从 Q-Learning 到 GRPO：强化学习算法、网络角色与部署选择</title><link>https://example.com/blog/rl-algorithms-qlearning-to-grpo/</link><guid isPermaLink="true">https://example.com/blog/rl-algorithms-qlearning-to-grpo/</guid><description>梳理 Q-Learning、DQN、PPO、SAC 与 GRPO，分清策略、价值、奖励和参考模型的职责，以及训练时谁更新、部署时保留谁。</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>强化学习</category><category>强化学习</category><category>Q-Learning</category><category>DQN</category><category>PPO</category><category>SAC</category><category>GRPO</category><category>RLHF</category></item><item><title>流式语音识别中的 Chunk 分块掩码：Conformer、Emformer、Zipformer 与 U2/U2++</title><link>https://example.com/blog/streaming-asr-chunk-masking/</link><guid isPermaLink="true">https://example.com/blog/streaming-asr-chunk-masking/</guid><description>从 Chunk-based Masking 出发，理解流式语音识别如何限制未来信息、复用历史缓存，并比较 Conformer、Emformer、U2/U2++ 和 Zipformer 的延迟与精度取舍。</description><pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>语音识别</category><category>流式语音识别</category><category>Streaming ASR</category><category>Conformer</category><category>Emformer</category><category>Zipformer</category><category>U2</category><category>语音识别</category></item><item><title>一次 Cloudflare Pages 构建失败排查：从 npm ci 锁文件报错到手机样式恢复</title><link>https://example.com/blog/cloudflare-pages-npm-ci-lockfile/</link><guid isPermaLink="true">https://example.com/blog/cloudflare-pages-npm-ci-lockfile/</guid><description>记录一次 QQ、百度手机浏览器页面样式异常的排查过程：从 Tailwind CSS 兼容性、Cloudflare 构建日志，到 npm ci 与 package-lock.json 不同步的修复和线上验证。</description><pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>网站搭建与部署</category><category>Cloudflare Pages</category><category>npm ci</category><category>package-lock.json</category><category>Astro</category><category>Tailwind CSS</category><category>网站部署</category><category>故障排查</category></item><item><title>ZeRO 显存优化：ZeRO-1、ZeRO-2 与 ZeRO-3</title><link>https://example.com/blog/zero-memory-optimization/</link><guid isPermaLink="true">https://example.com/blog/zero-memory-optimization/</guid><description>从混合精度 Adam 的显存账本出发，拆解 ZeRO 三阶段如何逐步切分优化器状态、梯度和参数，并用 7B、13B 例子估算显存与通信代价。</description><pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>微调与显存优化</category><category>大模型训练</category><category>显存优化</category><category>ZeRO</category><category>DeepSpeed</category><category>分布式训练</category><category>模型微调</category></item><item><title>笔记：用 Git Graph 回退当前进度并同步到远端</title><link>https://example.com/notes/git-graph-reset-and-sync/</link><guid isPermaLink="true">https://example.com/notes/git-graph-reset-and-sync/</guid><description>记录在 VS Code 的 Git Graph 中回退当前分支，并将回退结果安全同步到远端的图形化操作流程。</description><pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Git</category><category>Git Graph</category><category>版本回退</category><category>远端同步</category></item><item><title>随笔：用包容的眼光看待世界</title><link>https://example.com/essays/three-views-and-friendship/</link><guid isPermaLink="true">https://example.com/essays/three-views-and-friendship/</guid><description>从世界观、人生观和价值观出发，重新理解对错、立场、关系与边界：你的三观可以是地图，但不必成为衡量所有人的尺子。</description><pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate><category>三观</category><category>人际关系</category><category>认知边界</category><category>道德判断</category><category>庄子</category></item><item><title>笔记：OpenAI 如何让纳维尔–斯托克斯方程“爆炸”？</title><link>https://example.com/notes/openai-navier-stokes-blowup/</link><guid isPermaLink="true">https://example.com/notes/openai-navier-stokes-blowup/</guid><description>从速度场、不可压缩条件和非线性放大出发，理解 OpenAI 公开的受迫三维 Navier–Stokes 有限时间奇点构造，以及它对数学和工程意味着什么。</description><pubDate>Sun, 27 Sep 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>Navier–Stokes</category><category>流体力学</category><category>AI for Science</category><category>偏微分方程</category><category>奇点</category></item><item><title>LLaVA 是什么：从视觉指令微调到开源多模态生态</title><link>https://example.com/blog/llava-visual-instruction-tuning/</link><guid isPermaLink="true">https://example.com/blog/llava-visual-instruction-tuning/</guid><description>从 LLaVA 的视觉编码器、投影层和语言模型，到两阶段训练、GQA 数据集、LLaVA-1.5 以及它对后续开源多模态模型的影响。</description><pubDate>Sun, 27 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉语言模型</category><category>LLaVA</category><category>多模态模型</category><category>视觉语言模型</category><category>视觉指令微调</category><category>GQA</category></item><item><title>模型参数量怎么算：CNN、Transformer、ViT、LoRA 与 Adapter</title><link>https://example.com/blog/model-parameter-counting/</link><guid isPermaLink="true">https://example.com/blog/model-parameter-counting/</guid><description>从卷积核、注意力投影和前馈网络，到 ViT 的 patch embedding、LoRA 与 Adapter，掌握主流模型参数量的估算方法。</description><pubDate>Sun, 27 Sep 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>模型基础与架构</category><category>模型参数量</category><category>CNN</category><category>Transformer</category><category>ViT</category><category>LoRA</category><category>Adapter</category></item><item><title>视觉特征融合的五种经典玩法：FPN、PANet、BiFPN、ASPP 与 Non-local</title><link>https://example.com/blog/feature-fusion-methods/</link><guid isPermaLink="true">https://example.com/blog/feature-fusion-methods/</guid><description>从跨层金字塔、双向路径、可学习加权，到层内多尺度上下文和全局依赖，理解五种经典视觉特征融合模块的差异与适用边界。</description><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉理解</category><category>计算机视觉</category><category>特征融合</category><category>目标检测</category><category>语义分割</category><category>FPN</category><category>BiFPN</category></item><item><title>视觉 Token 压缩：ToMe、FastV、EViT 与 DiffRate</title><link>https://example.com/blog/visual-token-compression/</link><guid isPermaLink="true">https://example.com/blog/visual-token-compression/</guid><description>从重要性剪枝、相似度合并到可学习压缩率，理解视觉 token 压缩的位置、训练要求，以及 FLOPs 降低为何不等于推理同比加速。</description><pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉语言模型</category><category>多模态</category><category>视觉Token</category><category>模型加速</category><category>ToMe</category><category>FastV</category><category>EViT</category><category>DiffRate</category></item><item><title>VLA 的动作生成：离散 Token、扩散与 Flow Matching</title><link>https://example.com/blog/vla-action-generation/</link><guid isPermaLink="true">https://example.com/blog/vla-action-generation/</guid><description>从 OpenVLA、Octo 到 π0，理解动作量化、连续生成、动作分块，以及数据适配、微调和真实机器人部署中的关键边界。</description><pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate><category>具身智能与世界模型</category><category>动作生成</category><category>具身智能</category><category>VLA</category><category>OpenVLA</category><category>Octo</category><category>Flow Matching</category><category>机器人学习</category></item><item><title>多模态模型的视觉前端与对齐策略：CLIP、SigLIP、DINOv2、Q-Former 与 MLP</title><link>https://example.com/blog/multimodal-vision-encoders-and-connectors/</link><guid isPermaLink="true">https://example.com/blog/multimodal-vision-encoders-and-connectors/</guid><description>从视觉编码器到跨模态连接器，梳理图文预训练、视觉特征融合与语言模型对齐，并延伸到 VLA 的离散动作和连续动作生成。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉语言模型</category><category>多模态模型</category><category>CLIP</category><category>SigLIP</category><category>DINOv2</category><category>Q-Former</category><category>VLA</category></item><item><title>ONNX 和 TensorRT：模型部署加速的 1 分钟速通</title><link>https://example.com/blog/onnx-tensorrt-inference-acceleration/</link><guid isPermaLink="true">https://example.com/blog/onnx-tensorrt-inference-acceleration/</guid><description>从 PyTorch 导出 ONNX，到在 NVIDIA GPU 上构建 TensorRT engine，梳理模型转换、图优化、精度选择、动态形状与结果验证的完整链路。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>部署与推理加速</category><category>模型部署</category><category>ONNX</category><category>TensorRT</category><category>推理加速</category><category>PyTorch</category><category>Jetson</category></item><item><title>视频时序建模：从 C3D、I3D、TSM 到 TimeSformer 与 VideoMAE</title><link>https://example.com/blog/video-temporal-modeling/</link><guid isPermaLink="true">https://example.com/blog/video-temporal-modeling/</guid><description>从三维卷积、权重膨胀、时序移位到分解注意力和掩码预训练，理解五种经典视频建模方法的机制、边界与工程取舍。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉理解</category><category>视频理解</category><category>时序建模</category><category>C3D</category><category>I3D</category><category>TSM</category><category>TimeSformer</category><category>VideoMAE</category></item><item><title>视觉位置编码：RoPE、ALiBi 与 2D-RoPE 的 1 分钟速通</title><link>https://example.com/blog/visual-position-encoding-rope-alibi-2d-rope/</link><guid isPermaLink="true">https://example.com/blog/visual-position-encoding-rope-alibi-2d-rope/</guid><description>从 Transformer 为什么需要位置，到 RoPE 的旋转、ALiBi 的线性偏置和视觉 2D-RoPE 的行列坐标，理解图像与多模态模型如何知道 token 在哪里。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉语言模型</category><category>视觉位置编码</category><category>RoPE</category><category>ALiBi</category><category>2D-RoPE</category><category>Vision Transformer</category><category>多模态模型</category></item><item><title>世界模型的三条路线：Dreamer、Genie 与 JEPA</title><link>https://example.com/blog/world-model-routes-dreamer-genie-jepa/</link><guid isPermaLink="true">https://example.com/blog/world-model-routes-dreamer-genie-jepa/</guid><description>从潜空间想象、像素级生成到抽象表征预测，理解世界模型如何预测未来、学习动作与支持规划。</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>具身智能与世界模型</category><category>世界模型</category><category>世界模型</category><category>Dreamer</category><category>Genie</category><category>JEPA</category><category>具身智能</category><category>强化学习</category><category>多模态模型</category></item><item><title>具身智能的数据从哪里来？遥操作、仿真、UMI 与 Ego 数据采集全解</title><link>https://example.com/blog/embodied-ai-data-collection/</link><guid isPermaLink="true">https://example.com/blog/embodied-ai-data-collection/</guid><description>从机器人遥操作、仿真合成、UMI 手持夹爪到 Ego 第一视角数据，梳理具身智能数据采集的技术演变、适用场景与主要取舍。</description><pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate><category>具身智能与世界模型</category><category>数据采集</category><category>AI</category><category>具身智能</category><category>机器人</category><category>VLA</category><category>数据采集</category></item><item><title>机器人学习与最优控制：从 LQR、MPC 到 Isaac Sim、MuJoCo、SAPIEN 与 PyBullet</title><link>https://example.com/blog/robot-learning-optimal-control-simulators/</link><guid isPermaLink="true">https://example.com/blog/robot-learning-optimal-control-simulators/</guid><description>梳理 PID、LQR、iLQR、MPC、模仿学习与强化学习的关系，并比较 Isaac Sim/Isaac Lab、MuJoCo、SAPIEN 和 PyBullet 的定位、优势与适用场景。</description><pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate><category>具身智能与世界模型</category><category>控制与仿真</category><category>机器人学习</category><category>最优控制</category><category>仿真</category><category>Isaac Sim</category><category>MuJoCo</category><category>SAPIEN</category><category>PyBullet</category></item><item><title>从像素变换到世界模型：多模态数据增强的三层演进</title><link>https://example.com/blog/multimodal-data-augmentation/</link><guid isPermaLink="true">https://example.com/blog/multimodal-data-augmentation/</guid><description>以 SimCLR、VACoT、MIDAS、VISTA 和 Cosmos-Drive-Dreams 为线索，理解数据增强如何从改变像素发展到保持模态关系、扩展长视频并生成稀有场景。</description><pubDate>Tue, 15 Sep 2026 00:00:00 GMT</pubDate><category>多模态与视觉</category><category>视觉语言模型</category><category>AI</category><category>多模态学习</category><category>数据增强</category><category>VLM</category><category>VLA</category></item><item><title>随笔：vibecoding的小总结</title><link>https://example.com/essays/high-intensity-vibecoding-year-review/</link><guid isPermaLink="true">https://example.com/essays/high-intensity-vibecoding-year-review/</guid><description>回顾高强度使用 AI 辅助开发一年后的实践：如何完善环境、组织项目、调试协作，以及在 AI 时代保持清醒的工程心态。</description><pubDate>Sun, 30 Aug 2026 00:00:00 GMT</pubDate><category>Vibe Coding</category><category>AI 开发</category><category>工程实践</category><category>开发方法</category></item><item><title>随笔：当 AI 写完代码之后，还剩下什么？</title><link>https://example.com/essays/ai-code-value-reflection/</link><guid isPermaLink="true">https://example.com/essays/ai-code-value-reflection/</guid><description>当代码逐渐失去稀缺性，工程师的价值将从写出代码转向定义问题、理解业务与系统，并对最终结果负责。</description><pubDate>Thu, 27 Aug 2026 00:00:00 GMT</pubDate><category>AI</category><category>软件工程</category><category>工程实践</category><category>职业思考</category></item><item><title>不懂代码，也能看懂的个人网站搭建指南</title><link>https://example.com/blog/personal-website-tech-selection/</link><guid isPermaLink="true">https://example.com/blog/personal-website-tech-selection/</guid><description>把 Astro、Next.js、Markdown、Tailwind CSS、SSG 和托管平台这些名词分别放回它们负责的位置，再给出零基础做个人网站的选型建议和一份可以直接复制的提示词。</description><pubDate>Wed, 26 Aug 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>网站搭建与部署</category><category>个人网站</category><category>Astro</category><category>技术选型</category><category>静态网站</category></item><item><title>随笔：吉隆坡旅行观察日记</title><link>https://example.com/essays/kuala-lumpur-travel-notes/</link><guid isPermaLink="true">https://example.com/essays/kuala-lumpur-travel-notes/</guid><description>第一次出国去了吉隆坡：语言、城市建设、饮食、支付、住宅与楼层命名，记下一些不适应的地方和有意思的细节。</description><pubDate>Wed, 26 Aug 2026 00:00:00 GMT</pubDate><category>旅行</category><category>马来西亚</category><category>吉隆坡</category><category>观察日记</category></item><item><title>从训练模型到端侧部署：.ckpt、.pt、ONNX、PNNX、NCNN 与 2026 年的移动端推理全景</title><link>https://example.com/blog/model-deployment-guide/</link><guid isPermaLink="true">https://example.com/blog/model-deployment-guide/</guid><description>把模型格式、转换工具、推理运行时和硬件后端这四层彻底拆开，配一个能真正跑通的 ResNet-18 端到端例子，并覆盖量化、数值验证、Android 工程细节、端侧大模型与 NNAPI 落幕后的新格局。</description><pubDate>Tue, 25 Aug 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>部署与推理加速</category><category>模型部署</category><category>NCNN</category><category>ONNX</category><category>PNNX</category><category>Android</category><category>端侧推理</category><category>量化</category><category>PyTorch</category></item><item><title>为什么总有人极力推荐使用命令行操作而非图形界面</title><link>https://example.com/blog/command-line-vs-gui/</link><guid isPermaLink="true">https://example.com/blog/command-line-vs-gui/</guid><description>从可记录、可复现、可组合和远程协作等角度，理解命令行相对于图形界面的价值，也看见图形界面的优势与命令行的风险。</description><pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate><category>开发与工具实践</category><category>效率工具</category><category>命令行</category><category>效率工具</category><category>科研实践</category><category>图形界面</category></item><item><title>如何理解深度学习中的框架</title><link>https://example.com/blog/understanding-deep-learning-frameworks/</link><guid isPermaLink="true">https://example.com/blog/understanding-deep-learning-frameworks/</guid><description>从模型架构、训练流程和工具封装三个层面，理解什么是深度学习框架，以及用框架跑任务意味着什么。</description><pubDate>Thu, 13 Aug 2026 00:00:00 GMT</pubDate><category>大模型与训练优化</category><category>模型基础与架构</category><category>深度学习</category><category>PyTorch</category><category>工程实践</category></item><item><title>笔记：《深度学习入门》读书笔记</title><link>https://example.com/notes/deep-learning-from-scratch-notes/</link><guid isPermaLink="true">https://example.com/notes/deep-learning-from-scratch-notes/</guid><description>围绕神经网络学习、反向传播、优化技巧与卷积神经网络，整理《深度学习入门》的核心知识。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>读书笔记</category><category>深度学习</category><category>反向传播</category><category>优化器</category><category>卷积神经网络</category></item><item><title>笔记：Git 与 GitHub 核心概念及用法</title><link>https://example.com/notes/git-github-core-concepts/</link><guid isPermaLink="true">https://example.com/notes/git-github-core-concepts/</guid><description>从仓库、提交、分支与 HEAD 出发，整理远程协作、Pull Request、Rebase、Stash 等 Git 实践。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Git</category><category>GitHub</category><category>版本控制</category><category>协作开发</category></item><item><title>笔记：Mac Homebrew 使用指南</title><link>https://example.com/notes/mac-homebrew-guide/</link><guid isPermaLink="true">https://example.com/notes/mac-homebrew-guide/</guid><description>记录 Homebrew 的安装、PATH 配置、镜像设置，以及软件搜索、安装、更新和卸载方法。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Homebrew</category><category>macOS</category><category>包管理</category><category>命令行</category></item><item><title>笔记：用 Marvis 把旧电脑改造成家庭 AI 服务器</title><link>https://example.com/notes/marvis-home-ai-server/</link><guid isPermaLink="true">https://example.com/notes/marvis-home-ai-server/</guid><description>记录使用 Marvis 配置 Windows、开发环境、Ollama 与本地多模态模型的完整思路。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>Marvis</category><category>家庭服务器</category><category>Ollama</category><category>本地模型</category></item><item><title>笔记：Matplotlib 实用教程</title><link>https://example.com/notes/matplotlib-practical-guide/</link><guid isPermaLink="true">https://example.com/notes/matplotlib-practical-guide/</guid><description>整理 Matplotlib 的基础绘图流程，以及图形配置、坐标轴、图例和布局工具的使用方式。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>课程笔记</category><category>Matplotlib</category><category>Python</category><category>数据可视化</category><category>绘图</category></item><item><title>笔记：Python 数据科学：Pandas 与数据处理</title><link>https://example.com/notes/python-data-science-notes/</link><guid isPermaLink="true">https://example.com/notes/python-data-science-notes/</guid><description>整理 Pandas 的 Series、DataFrame、索引、缺失值、分组聚合与常用数据处理方法。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>课程笔记</category><category>Python</category><category>Pandas</category><category>NumPy</category><category>数据分析</category></item><item><title>笔记：小土堆 PyTorch 课程笔记</title><link>https://example.com/notes/pytorch-course-notes/</link><guid isPermaLink="true">https://example.com/notes/pytorch-course-notes/</guid><description>从数据集、TensorBoard 与图像变换开始，整理 PyTorch 网络搭建、训练、保存和 GPU 使用流程。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>课程笔记</category><category>PyTorch</category><category>TensorBoard</category><category>神经网络</category><category>模型训练</category></item><item><title>笔记：uv Python 项目管理指南</title><link>https://example.com/notes/uv-python-project-guide/</link><guid isPermaLink="true">https://example.com/notes/uv-python-project-guide/</guid><description>记录 uv 的安装、项目初始化、依赖添加、PyTorch 索引配置和环境复用方法。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>uv</category><category>Python</category><category>依赖管理</category><category>虚拟环境</category></item><item><title>笔记：Vibe Coding：5 招破解调试黑箱</title><link>https://example.com/notes/vibe-coding-debugging-guide/</link><guid isPermaLink="true">https://example.com/notes/vibe-coding-debugging-guide/</guid><description>从理解设计、增加日志、测试驱动、Git 存档和主动复述五个角度，整理 AI 编程调试方法。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>Vibe Coding</category><category>调试</category><category>测试</category><category>Git</category></item><item><title>笔记：Web of Science 检索要点</title><link>https://example.com/notes/web-of-science-search-guide/</link><guid isPermaLink="true">https://example.com/notes/web-of-science-search-guide/</guid><description>记录逻辑运算符、通配符、短语和作者检索，以及提高文献查全率与查准率的方法。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Web of Science</category><category>文献检索</category><category>科研工具</category><category>检索式</category></item><item><title>笔记：Word 排版与论文写作笔记</title><link>https://example.com/notes/word-formatting-and-writing-notes/</link><guid isPermaLink="true">https://example.com/notes/word-formatting-and-writing-notes/</guid><description>整理 Word 段落、样式、表格、题注、页眉页脚、参考文献与长文档管理中的常见操作。</description><pubDate>Sun, 19 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>Word</category><category>论文写作</category><category>文档排版</category><category>长文档</category></item><item><title>笔记：大 A 限定黑话：做 T 到底是什么？如何做？</title><link>https://example.com/notes/a-share-doing-t/</link><guid isPermaLink="true">https://example.com/notes/a-share-doing-t/</guid><description>解释 A 股语境中的“做 T”、底仓与 T+1 约束，并整理常见操作方式和新手误区。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>A股</category><category>做T</category><category>交易机制</category><category>风险管理</category></item><item><title>笔记：激活函数演进史：从 Sigmoid 到 SwiGLU</title><link>https://example.com/notes/activation-functions-evolution/</link><guid isPermaLink="true">https://example.com/notes/activation-functions-evolution/</guid><description>沿着 Sigmoid、Tanh、ReLU、SiLU、GeLU 到 SwiGLU，理解激活函数如何围绕梯度传播与门控机制演进。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>课程笔记</category><category>激活函数</category><category>深度学习</category><category>SwiGLU</category><category>神经网络</category></item><item><title>笔记：K 线图完全指南：从起源到实战解读</title><link>https://example.com/notes/candlestick-chart-guide/</link><guid isPermaLink="true">https://example.com/notes/candlestick-chart-guide/</guid><description>从开高低收四个价格出发，理解 K 线的结构、颜色惯例、常见形态和阅读时的风险边界。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>K线</category><category>蜡烛图</category><category>技术分析</category><category>金融基础</category></item><item><title>笔记：杠杆进阶：专业机构如何使用杠杆</title><link>https://example.com/notes/institutional-leverage-basics/</link><guid isPermaLink="true">https://example.com/notes/institutional-leverage-basics/</guid><description>从投资暴露、强制平仓、波动率和夏普比率出发，理解杠杆如何同时放大收益、风险与融资成本。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>杠杆</category><category>风险管理</category><category>夏普比率</category><category>资产配置</category></item><item><title>笔记：马丁策略与反马丁策略：仓位管理解析</title><link>https://example.com/notes/martingale-vs-antimartingale/</link><guid isPermaLink="true">https://example.com/notes/martingale-vs-antimartingale/</guid><description>比较马丁与反马丁两种仓位管理思路，说明加仓、止损、趋势跟踪与资金承受能力之间的关系。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>仓位管理</category><category>马丁策略</category><category>反马丁策略</category><category>风险管理</category></item><item><title>笔记：Qwen3-VL-Seg 全文精读</title><link>https://example.com/notes/qwen3-vl-seg-paper-reading/</link><guid isPermaLink="true">https://example.com/notes/qwen3-vl-seg-paper-reading/</guid><description>按原文、翻译和通俗解释三层结构，梳理 Qwen3-VL-Seg 的开放世界指代分割方法、数据集、训练方案与实验结果。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>读书笔记</category><category>Qwen3-VL-Seg</category><category>多模态</category><category>指代分割</category><category>论文精读</category></item><item><title>笔记：RLHF 由浅入深全面解析</title><link>https://example.com/notes/rlhf-from-basics-to-practice/</link><guid isPermaLink="true">https://example.com/notes/rlhf-from-basics-to-practice/</guid><description>从 SFT、奖励模型与 PPO 三阶段出发，梳理 RLHF 的训练目标、关键公式、常见变体与实践挑战。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>课程笔记</category><category>RLHF</category><category>SFT</category><category>PPO</category><category>DPO</category><category>大语言模型</category></item><item><title>笔记：SVG 矢量图：从基础代码到实战应用</title><link>https://example.com/notes/svg-from-basics-to-practice/</link><guid isPermaLink="true">https://example.com/notes/svg-from-basics-to-practice/</guid><description>系统整理 SVG 的坐标系、基础形状、路径、样式、复用、变换、动画和 Web 集成方式。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>SVG</category><category>矢量图</category><category>Web</category><category>图形基础</category></item><item><title>笔记：WMF 矢量图：从 Windows 元文件到实战应用</title><link>https://example.com/notes/wmf-from-basics-to-practice/</link><guid isPermaLink="true">https://example.com/notes/wmf-from-basics-to-practice/</guid><description>梳理 WMF/EMF 文件结构、绘图记录、坐标映射、位图操作、Windows API 与格式转换。</description><pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate><category>工具记录</category><category>WMF</category><category>EMF</category><category>矢量图</category><category>Windows API</category></item><item><title>笔记：上市公司科普知识</title><link>https://example.com/notes/listed-company-basics/</link><guid isPermaLink="true">https://example.com/notes/listed-company-basics/</guid><description>从 IPO、资金流向和股权稀释出发，梳理上市融资、一级与二级市场，以及企业选择上市的利弊。</description><pubDate>Thu, 16 Jul 2026 00:00:00 GMT</pubDate><category>随手整理</category><category>IPO</category><category>上市公司</category><category>股权融资</category><category>金融常识</category></item></channel></rss>