Tags
Bolderdash
Toggle navigation
Jarxi
Home
中文
About
Technical
Elastic Beanstalk
AI
Architecture
React
hexo
ubuntu
Node
Angular
Docker
OpenClaw
SFT
Tokenizer
GLM
AWS
Amazon
Review
C
DraftJS
ES6
Spring boot
Gradle
learning
Flask
Linux
JWT
Security
Operating System
Memory
Python
RBG
Regex
胡言乱语
Startup
Agent
VR
HuggingFace
github
TypeScript
Data
System Design
Queue
Auth
Monorepo
Database
Postgres
Billing
Prometheus
Monitoring
Networking
Reddit
Startups
Marketing
Agent Ops
Inference
MTP
dual boot
tools
Reasoning
Claude Code
Writing
Plugin
Technical
One API Client, Two Authentication Channels
The web uses an HTTP-only cookie, the CLI uses a bearer token. The server checks both in one middleware, and each shell injects its own way of producing credentials — so feature code never learns which platform it is on.
Next.js Is Just a Shell, Peer to Electron
The easiest mistake in a multi-client monorepo is treating the Next.js app as the product and everything else as a port. Demote it to a shell and the package boundaries fall out — especially the one the CLI forces on you.
Cross-Package Imports Are Type-Only, Runtime Goes Through ctx
import type compiles to nothing, so it creates no runtime dependency edge. That is how dozens of UI packages can know each other's types while never loading each other — the plugin boundary lives in one keyword.
Judge Your Judge Twice Before You Trust It
I used a large model as a judge to fact-check SFT data. Judging the same note twice gave zero errors on one run and a dozen on the next. A single verdict is a coin flip.
Prometheus Behind NAT: Scraping Without Open Ports
How a pull-based monitor reaches a machine with no public IP. The tunnel is a phone line, not a mailbox — and nothing exists until someone asks for it.
Load Leveling vs Load Balancing: The Fan-Out Pattern
Two queues, one event. Load balancing splits work across identical workers. Load leveling spreads work across time, so a slow consumer never blocks a fast one.
Speculative Decoding and MTP: Why Guessing Is Free
A forward pass over five tokens costs about the same as over one. That gap is the entire speedup, and MTP is how the model drafts guesses to fill it.
What special: true Actually Changes
It changes nothing when you tokenize. It changes everything when you detokenize, and only if skip_special_tokens is on. Measured against GLM 5.2's tokenizer.
After the Hexo Skill, a Plugin for Tone
My first Claude Code skill has 700+ downloads on ClawHub, which surprised me enough to write another. This one holds the voices I post in, and it will keep growing.
What Actually Gets Tokenized in SFT
The model never sees your messages list, only the rendered string. Once that clicks, add_generation_prompt, loss masks and the double-BOS bug all line up behind it.
Multi-level Account and User Design: One User, One Account
A three-level account and user model — legal entity, account, user. Why we rejected this design, then shipped it two months later when one rule changed.
Using the blog-hexo skill
Draft + publish workflow in one place
Elastic Beanstalk
[Elastic Beanstalk] Springboot Gradle Docker Elastic Beanstalk
[Elastic Beanstalk] Deploy Angular & Node to Elastic Beanstalk
[Elastic Beanstalk] Elastic Beanstalk Commands
[Hexo] Deploy Hexo to Elastic Beanstalk
[Elastic Beanstalk] React and Node to Elastic Beanstalk
AI
Judge Your Judge Twice Before You Trust It
I used a large model as a judge to fact-check SFT data. Judging the same note twice gave zero errors on one run and a dozen on the next. A single verdict is a coin flip.
Speculative Decoding and MTP: Why Guessing Is Free
A forward pass over five tokens costs about the same as over one. That gap is the entire speedup, and MTP is how the model drafts guesses to fill it.
What special: true Actually Changes
It changes nothing when you tokenize. It changes everything when you detokenize, and only if skip_special_tokens is on. Measured against GLM 5.2's tokenizer.
After the Hexo Skill, a Plugin for Tone
My first Claude Code skill has 700+ downloads on ClawHub, which surprised me enough to write another. This one holds the voices I post in, and it will keep growing.
What Actually Gets Tokenized in SFT
The model never sees your messages list, only the rendered string. Once that clicks, add_generation_prompt, loss masks and the double-BOS bug all line up behind it.
Architecture
One API Client, Two Authentication Channels
The web uses an HTTP-only cookie, the CLI uses a bearer token. The server checks both in one middleware, and each shell injects its own way of producing credentials — so feature code never learns which platform it is on.
Next.js Is Just a Shell, Peer to Electron
The easiest mistake in a multi-client monorepo is treating the Next.js app as the product and everything else as a port. Demote it to a shell and the package boundaries fall out — especially the one the CLI forces on you.
Cross-Package Imports Are Type-Only, Runtime Goes Through ctx
import type compiles to nothing, so it creates no runtime dependency edge. That is how dozens of UI packages can know each other's types while never loading each other — the plugin boundary lives in one keyword.
Load Leveling vs Load Balancing: The Fan-Out Pattern
Two queues, one event. Load balancing splits work across identical workers. Load leveling spreads work across time, so a slow consumer never blocks a fast one.
Multi-level Account and User Design: One User, One Account
A three-level account and user model — legal entity, account, user. Why we rejected this design, then shipped it two months later when one rule changed.
React
[MEAN] How to use httpOnly JWT with React and Node
[React] React Notes
[DraftJS] DraftJS RichText and Plugins
[Elastic Beanstalk] React and Node to Elastic Beanstalk
hexo
Using the blog-hexo skill
Draft + publish workflow in one place
[Hexo] Deploy Hexo to Elastic Beanstalk
[Hexo] How to Use My Hexo
[Hexo] Hexo commands
ubuntu
[ubuntu] Bluetooth stopped working after changing /etc/bluetooth/main.config
[ubuntu] Linux install Netease
[ubuntu] Windows & Ubuntu dual boot e.g. Aurora R8
[ubuntu] MySQL outfile
Node
[MEAN] How to use httpOnly JWT with React and Node
[Elastic Beanstalk] Deploy Angular & Node to Elastic Beanstalk
[Elastic Beanstalk] React and Node to Elastic Beanstalk
Angular
[Elastic Beanstalk] Deploy Angular & Node to Elastic Beanstalk
[Angular] Autocomplete
Docker
[Docker] Docker For Beginners
[Elastic Beanstalk] Springboot Gradle Docker Elastic Beanstalk
OpenClaw
Using the blog-hexo skill
Draft + publish workflow in one place
Organization of OpenClaw Memory
SFT
Judge Your Judge Twice Before You Trust It
I used a large model as a judge to fact-check SFT data. Judging the same note twice gave zero errors on one run and a dozen on the next. A single verdict is a coin flip.
What Actually Gets Tokenized in SFT
The model never sees your messages list, only the rendered string. Once that clicks, add_generation_prompt, loss masks and the double-BOS bug all line up behind it.
Tokenizer
What special: true Actually Changes
It changes nothing when you tokenize. It changes everything when you detokenize, and only if skip_special_tokens is on. Measured against GLM 5.2's tokenizer.
What Actually Gets Tokenized in SFT
The model never sees your messages list, only the rendered string. Once that clicks, add_generation_prompt, loss masks and the double-BOS bug all line up behind it.
GLM
Speculative Decoding and MTP: Why Guessing Is Free
A forward pass over five tokens costs about the same as over one. That gap is the entire speedup, and MTP is how the model drafts guesses to fill it.
What special: true Actually Changes
It changes nothing when you tokenize. It changes everything when you detokenize, and only if skip_special_tokens is on. Measured against GLM 5.2's tokenizer.
AWS
[AWS] Point Route 53 to Github Page
Amazon
[Amazon] How to update brand name
Review
[Book] Anything You Want: 40 Lessons for a New Kind of Entrepreneur
C
[C] Initalized struct
DraftJS
[DraftJS] DraftJS RichText and Plugins
ES6
[ES6] ES6 Syntax
Spring boot
[Elastic Beanstalk] Springboot Gradle Docker Elastic Beanstalk
Gradle
[Elastic Beanstalk] Springboot Gradle Docker Elastic Beanstalk
learning
[Team] Eslint Prettier Setup
Flask
[Flask] Flask Tutorial
Linux
[Linux] Linux File System
JWT
[MEAN] How to use httpOnly JWT with React and Node
Security
[MEAN] How to use httpOnly JWT with React and Node
Operating System
[Operating System] Spinlock
Memory
Organization of OpenClaw Memory
Python
[Python] How to Setup Python Env
RBG
[RBG] The thoughts after RBG's death
Regex
[Regex] Regular Expression Tutorial
胡言乱语
[胡言乱语] Bridging the gap between experts and decision makers
Startup
[Startup] Trademark Registration
Agent
Using the blog-hexo skill
Draft + publish workflow in one place
VR
[VR] How to connect Oculus Rift to Unity
HuggingFace
What Actually Gets Tokenized in SFT
The model never sees your messages list, only the rendered string. Once that clicks, add_generation_prompt, loss masks and the double-BOS bug all line up behind it.
github
[github] github commands habits
TypeScript
Cross-Package Imports Are Type-Only, Runtime Goes Through ctx
import type compiles to nothing, so it creates no runtime dependency edge. That is how dozens of UI packages can know each other's types while never loading each other — the plugin boundary lives in one keyword.
Data
Judge Your Judge Twice Before You Trust It
I used a large model as a judge to fact-check SFT data. Judging the same note twice gave zero errors on one run and a dozen on the next. A single verdict is a coin flip.
System Design
Load Leveling vs Load Balancing: The Fan-Out Pattern
Two queues, one event. Load balancing splits work across identical workers. Load leveling spreads work across time, so a slow consumer never blocks a fast one.
Queue
Load Leveling vs Load Balancing: The Fan-Out Pattern
Two queues, one event. Load balancing splits work across identical workers. Load leveling spreads work across time, so a slow consumer never blocks a fast one.
Auth
One API Client, Two Authentication Channels
The web uses an HTTP-only cookie, the CLI uses a bearer token. The server checks both in one middleware, and each shell injects its own way of producing credentials — so feature code never learns which platform it is on.
Monorepo
Next.js Is Just a Shell, Peer to Electron
The easiest mistake in a multi-client monorepo is treating the Next.js app as the product and everything else as a port. Demote it to a shell and the package boundaries fall out — especially the one the CLI forces on you.
Database
Multi-level Account and User Design: One User, One Account
A three-level account and user model — legal entity, account, user. Why we rejected this design, then shipped it two months later when one rule changed.
Postgres
Multi-level Account and User Design: One User, One Account
A three-level account and user model — legal entity, account, user. Why we rejected this design, then shipped it two months later when one rule changed.
Billing
Multi-level Account and User Design: One User, One Account
A three-level account and user model — legal entity, account, user. Why we rejected this design, then shipped it two months later when one rule changed.
Prometheus
Prometheus Behind NAT: Scraping Without Open Ports
How a pull-based monitor reaches a machine with no public IP. The tunnel is a phone line, not a mailbox — and nothing exists until someone asks for it.
Monitoring
Prometheus Behind NAT: Scraping Without Open Ports
How a pull-based monitor reaches a machine with no public IP. The tunnel is a phone line, not a mailbox — and nothing exists until someone asks for it.
Networking
Prometheus Behind NAT: Scraping Without Open Ports
How a pull-based monitor reaches a machine with no public IP. The tunnel is a phone line, not a mailbox — and nothing exists until someone asks for it.
Reddit
Reddit Growth Strategy: Warm-Up, Targeting, and Auto-Fix
How we keep daily Reddit ops on track without tripping moderator wires or API restrictions
Startups
Reddit Growth Strategy: Warm-Up, Targeting, and Auto-Fix
How we keep daily Reddit ops on track without tripping moderator wires or API restrictions
Marketing
Reddit Growth Strategy: Warm-Up, Targeting, and Auto-Fix
How we keep daily Reddit ops on track without tripping moderator wires or API restrictions
Agent Ops
Reddit Growth Strategy: Warm-Up, Targeting, and Auto-Fix
How we keep daily Reddit ops on track without tripping moderator wires or API restrictions
Inference
Speculative Decoding and MTP: Why Guessing Is Free
A forward pass over five tokens costs about the same as over one. That gap is the entire speedup, and MTP is how the model drafts guesses to fill it.
MTP
Speculative Decoding and MTP: Why Guessing Is Free
A forward pass over five tokens costs about the same as over one. That gap is the entire speedup, and MTP is how the model drafts guesses to fill it.
dual boot
[ubuntu] Windows & Ubuntu dual boot e.g. Aurora R8
tools
[VSCode] vscode-awesome-tools
Reasoning
What special: true Actually Changes
It changes nothing when you tokenize. It changes everything when you detokenize, and only if skip_special_tokens is on. Measured against GLM 5.2's tokenizer.
Claude Code
After the Hexo Skill, a Plugin for Tone
My first Claude Code skill has 700+ downloads on ClawHub, which surprised me enough to write another. This one holds the voices I post in, and it will keep growing.
Writing
After the Hexo Skill, a Plugin for Tone
My first Claude Code skill has 700+ downloads on ClawHub, which surprised me enough to write another. This one holds the voices I post in, and it will keep growing.
Plugin
After the Hexo Skill, a Plugin for Tone
My first Claude Code skill has 700+ downloads on ClawHub, which surprised me enough to write another. This one holds the voices I post in, and it will keep growing.