archive node
Categories
目录导航 + 文章分类归档
目录导航
Core Tracks
Systems & Infra
Tools & Engineering
Domain Notes
文章分类（自动）
AI
Inference
RAGFlow
agent
- MCP Configuration in VS Code and Copilot CLI
- Claude Code and GitHub Copilot CLI: Two Very Different Single-File Runtimes
- Claude Skills in VS Code Copilot and Copilot CLI
- Claude Code, GitHub Copilot, and Cline: Different Layers of AI-Assisted Development
- How Coding Agents Use System Prompts and Tools
agents
ai
- Why a 27B Model Runs at 5 tok/s on CPU: llama.cpp, vLLM, and GPUs
- Tokenizer
- Intro to Large Language Models
- Training a Tokenizer with train_new_from_iterator
- 人工智能、机器学习与自然语言处理的起源与关系
- Thinking About Retraining Kimi K3
- OpenAI vs LiteLLM: API Cheat Sheet
- Why Claude Code and Copilot CLI Choose the JavaScript Ecosystem
- MCP Configuration in VS Code and Copilot CLI
- Claude Code and GitHub Copilot CLI: Two Very Different Single-File Runtimes
- How an AI Agent Keeps Context Without Keeping Everything
- Claude Skills in VS Code Copilot and Copilot CLI
- Train Study
- Fixing CUDA Out-of-Memory Errors in Ovis2.5-2B Inference
- Claude Code, GitHub Copilot, and Cline: Different Layers of AI-Assisted Development
- How Coding Agents Use System Prompts and Tools
- Skills and MCP Config: Where VS Code and Copilot CLI Actually Look
- Claude Code Sessions: --continue, --resume, and Where Transcripts Live
- How Docker Compose Names RAGFlow's Containers Without `container_name`
- How RAGFlow's Docker Compose Decides Which Services Actually Start
- LiteLLM Proxy: One model_list, Both OpenAI and Anthropic Endpoints
- MCP 2026-07-28: Stateless Servers, Multi-Round-Trip Requests, and Safer Auth
- How to Enter the RAGFlow Admin Page
- How the RAGFlow Web UI Triggers link_connectors
- Optimizing a Bash .env Loader Without eval
- How RAGFlow Builds the Document Filter Panel
- How RAGFlow Syncs Data Sources and Builds Searchable Chunks
- RAGFlow Data Source Connectors: From Discovery to Chunks
- AI Practice 3 - Enabling RAGFlow HTTPS Through SSH
- Using AI to Expose Amazon Bedrock Claude Through an OpenAI-Compatible API
- Why RAGFlow Cannot List AWS Models When It Adds a Bearer Prefix
- Connecting RAGFlow to an Internal Confluence Instance
- Using AI for Repetitive Docker Debugging
- Understanding RAGFlow Chunk Tasks: Building, Embedding, and Storing Chunks
- How RAGFlow Renders SVG Icons from iconfont.js
- Why AI Helps More with Performance Optimization Than Feature Development
- How RAGFlow v0.26.4 Replaces TenantLLM with a Provider-Instance-Model Design
- How a RAGFlow Startup-Order Change Can Create Two Superusers
- Proxy Environment Variables Can Break RAGFlow MCP Services
- Reliable NO_PROXY Rules for RAGFlow MCP Clients in Docker
- RAGFlow's Default MySQL Host Port Changed from 5455 to 3306
- Why RAGFlow Adds the INVALID_ Prefix to Access Tokens
- Add a RAGFlow MCP Server to VS Code
- Rotary Position Embedding in GLM, DeepSeek-V3, and Qwen3
- Ragflow Superuser
- Understanding the Retriever Flow in RAGFlow
- Text Embeddings vs Visual Embeddings in Multimodal Transformers
- How to Analyze a Model from safetensors: Decoder Layers, Attention, and FFN
- Understanding PyTorch Scaled Dot-Product Attention
- Transformer Training Library Guide
- The Transformer: Introduction and History
- PyTorch: Evolution from Torch and the Deep Learning Framework Landscape
- Where to Find (Near) Real-Time Deep Learning Framework Market Share Data
- Transformer Training Pitfalls: OOM, Attention Complexity, and Getting the Objective Right
- Debugging RAGAS Evaluations: Exceptions, Tracking, and Concurrency Control
- Fixing "copilot is not recognized" on Windows
- Fine-Tuning with TRL: SFTTrainer, OOM, and LoRA
apps
- Why Claude Code and Copilot CLI Choose the JavaScript Ecosystem
- MCP Configuration in VS Code and Copilot CLI
- Claude Code and GitHub Copilot CLI: Two Very Different Single-File Runtimes
- Claude Skills in VS Code Copilot and Copilot CLI
- Claude Code, GitHub Copilot, and Cline: Different Layers of AI-Assisted Development
- How Coding Agents Use System Prompts and Tools
- How Docker Compose Names RAGFlow's Containers Without `container_name`
- How RAGFlow's Docker Compose Decides Which Services Actually Start
- How to Enter the RAGFlow Admin Page
- How the RAGFlow Web UI Triggers link_connectors
- How RAGFlow Builds the Document Filter Panel
- How RAGFlow Syncs Data Sources and Builds Searchable Chunks
- RAGFlow Data Source Connectors: From Discovery to Chunks
- Why RAGFlow Cannot List AWS Models When It Adds a Bearer Prefix
- Connecting RAGFlow to an Internal Confluence Instance
- Understanding RAGFlow Chunk Tasks: Building, Embedding, and Storing Chunks
- How RAGFlow Renders SVG Icons from iconfont.js
- How RAGFlow v0.26.4 Replaces TenantLLM with a Provider-Instance-Model Design
- How a RAGFlow Startup-Order Change Can Create Two Superusers
- Proxy Environment Variables Can Break RAGFlow MCP Services
- Reliable NO_PROXY Rules for RAGFlow MCP Clients in Docker
- RAGFlow's Default MySQL Host Port Changed from 5455 to 3306
- Why RAGFlow Adds the INVALID_ Prefix to Access Tokens
- Add a RAGFlow MCP Server to VS Code
- Ragflow Superuser
- Understanding the Retriever Flow in RAGFlow
- Debugging RAGAS Evaluations: Exceptions, Tracking, and Concurrency Control
- Fixing "copilot is not recognized" on Windows
architect
architecture
authentication
azure
bash
cache
certification
claude-code
computer-graphics
container
copilot
copilot-cli
cpu
- 从反射公式到 DDRIO 初始化代码：一套阅读方法
- 为什么需要 RCOMP、校准和 Training
- DB ODT、RDIMM 和 LRDIMM：先把两段链路画出来
- DDR5 读写方向变化：谁是源端，谁是负载端
- ODT、RON、RCOMP：三个名字分别在控制什么
- HBM: Trading Data Rate for Bus Width
- 从一次反射到振铃和眼图闭合
- 阻抗匹配到底匹配什么：源端、负载端和终端
- 5 V 为什么会出现 2.5 V：入射波、分压和开路瞬态
- 反射不是电压凭空变大：从边界条件推导反射系数
- 传输线上的 V+ 和 V-：输入输出之外的两个方向
- 什么时候一根导线必须当作传输线
- Who Are These DDR and DRAM Resources?
- 从电路节点到端口：电压、电流和输入输出不是一回事
- DQS, CTLE, and DFE in High-Speed DDR5 Receivers
- Why a Memory Platform May Raise RCD VDDIO from 1.0 V to 1.2 V
- Why CPU State Uses Flip-Flops, Latches, and SRAM Differently
- Cache Coherency, TLB Flush, and DMA: A Practical Memory-System Tour
- Inside Smartphone RAM: Decoding Markings, Die Stacking, and Why LPDDR Still Needs Training
database
devops
engineering
file-system
finance
frontend
github
gpu
hardware
identity
infer
jekyll
linux
- Running llama.cpp as a systemd Service on Debian
- Why a systemd-Enabled GitHub Actions Runner Still Shows Offline
- Bash Script Paths, Regex Captures, and .env Variable Expansion
- Is `while true; do ... done &` an Infinite Loop?
- Understanding ~/.bashrc, ~/.profile, /etc/bash.bashrc, and /etc/environment
- Why the Linux users Command Shows the Same User Multiple Times
- Understanding procps, top, and ps on Linux
- Why dpkg -S May Not Find /usr/bin/bash on Ubuntu
- Mainstream Linux Distributions: Debian, Ubuntu, CentOS, Fedora, and SUSE
- Understanding GNU Coreutils User and Group Commands
- Bash, Dash, and Common Shell Packages
- Upgrading CUDA Toolkit Without Touching the Driver: A Practical Guide
litellm
markdown
math
mcp
memory
- 从反射公式到 DDRIO 初始化代码：一套阅读方法
- 为什么需要 RCOMP、校准和 Training
- DB ODT、RDIMM 和 LRDIMM：先把两段链路画出来
- DDR5 读写方向变化：谁是源端，谁是负载端
- ODT、RON、RCOMP：三个名字分别在控制什么
- HBM: Trading Data Rate for Bus Width
- 从一次反射到振铃和眼图闭合
- 阻抗匹配到底匹配什么：源端、负载端和终端
- 5 V 为什么会出现 2.5 V：入射波、分压和开路瞬态
- 反射不是电压凭空变大：从边界条件推导反射系数
- 传输线上的 V+ 和 V-：输入输出之外的两个方向
- 什么时候一根导线必须当作传输线
- Who Are These DDR and DRAM Resources?
- 从电路节点到端口：电压、电流和输入输出不是一回事
- DQS, CTLE, and DFE in High-Speed DDR5 Receivers
- Why a Memory Platform May Raise RCD VDDIO from 1.0 V to 1.2 V
- Inside Smartphone RAM: Decoding Markings, Die Stacking, and Why LPDDR Still Needs Training
memory-management
microsoft
- How to Apply a PowerPoint Template to an Existing Presentation
- PowerShell Pipeline Cheat Sheet
- Choosing the Right ACL API in Azure
- Why a Python Service and an Interactive Shell Used Different Data Files
- Fix Microsoft Entra Sign-In Failures Caused by Required User Assignment
- PowerShell Command History: HistoryInfo and PSReadLine Persistence
- Windows PATH: Process, User, and Machine Scopes in PowerShell
- Authenticate Microsoft Graph Requests with Microsoft Entra ID
network
os
performance
powershell
programming
prompts
python
- Why a Python Service and an Interactive Shell Used Different Data Files
- Use the Windows Certificate Store with Python Requests
- Start a Visdom Server with Python
- Python HTTP Clients and TLS Certificate Verification
- Create Nested Directories in Python (os vs pathlib)
- Understanding Negative Lookahead in Regular Expressions
rag
- How Docker Compose Names RAGFlow's Containers Without `container_name`
- How RAGFlow's Docker Compose Decides Which Services Actually Start
- How to Enter the RAGFlow Admin Page
- Why RAGFlow Cannot List AWS Models When It Adds a Bearer Prefix
- Understanding RAGFlow Chunk Tasks: Building, Embedding, and Storing Chunks
- How RAGFlow Renders SVG Icons from iconfont.js
- How RAGFlow v0.26.4 Replaces TenantLLM with a Provider-Instance-Model Design
- How a RAGFlow Startup-Order Change Can Create Two Superusers
- Proxy Environment Variables Can Break RAGFlow MCP Services
- Reliable NO_PROXY Rules for RAGFlow MCP Clients in Docker
- RAGFlow's Default MySQL Host Port Changed from 5455 to 3306
- Why RAGFlow Adds the INVALID_ Prefix to Access Tokens
- Understanding the Retriever Flow in RAGFlow
ragas
ragflow
- How Docker Compose Names RAGFlow's Containers Without `container_name`
- How RAGFlow's Docker Compose Decides Which Services Actually Start
- How to Enter the RAGFlow Admin Page
- How the RAGFlow Web UI Triggers link_connectors
- How RAGFlow Builds the Document Filter Panel
- How RAGFlow Syncs Data Sources and Builds Searchable Chunks
- RAGFlow Data Source Connectors: From Discovery to Chunks
- Why RAGFlow Cannot List AWS Models When It Adds a Bearer Prefix
- Connecting RAGFlow to an Internal Confluence Instance
- Understanding RAGFlow Chunk Tasks: Building, Embedding, and Storing Chunks
- How RAGFlow Renders SVG Icons from iconfont.js
- How RAGFlow v0.26.4 Replaces TenantLLM with a Provider-Instance-Model Design
- How a RAGFlow Startup-Order Change Can Create Two Superusers
- Proxy Environment Variables Can Break RAGFlow MCP Services
- Reliable NO_PROXY Rules for RAGFlow MCP Clients in Docker
- RAGFlow's Default MySQL Host Port Changed from 5455 to 3306
- Why RAGFlow Adds the INVALID_ Prefix to Access Tokens
- Add a RAGFlow MCP Server to VS Code
- Ragflow Superuser
- Understanding the Retriever Flow in RAGFlow
regex
resources
security
- chmod vs. setfacl: Where Linux File Permissions Live
- PKI and Certificate Authorities: Who Actually Issues a Certificate?
- What the HTTP Authorization Header Actually Standardizes
- mkcert vs. OpenSSL for Local TLS Certificates
- Use the Windows Certificate Store with Python Requests
- Authenticate Microsoft Graph Requests with Microsoft Entra ID
- Inspecting Kerberos Tickets and Using Them with Python Requests
- Python HTTP Clients and TLS Certificate Verification
software
software-engineering
storage
tls
tools
train
virtualization
visdom
visualization
wandb
web
- Python Requests: `data=` and `json=` Are Not the Same
- Serving a Frontend Under an Nginx Subpath
- What the HTTP Authorization Header Actually Standardizes
- Web Frameworks
- Getting the Client's IP in FastAPI/Flask, and Why X-Real-IP Isn't There Locally
- Two Sites Every Frontend Developer Recognizes - Vercel and CSS-Tricks
- Why Tutorial Sites Like W3Schools Stay Profitable for Decades
- Electron, and What Tonsky's 'Claude is an Electron App' Post Actually Argues
- Where `li { display: list-item; unicode-bidi: isolate; }` Actually Comes From
- A Jekyll + kramdown Bug That Explains How CSS Selectors Actually Work
- Choosing a Cloud Service Provider for Domains and Hosting
- Why Chrome Automatically Redirects HTTP to HTTPS — and How to Bypass It