ついにBitNet Llama8Bが登場! CPUのみで爆速推論するLLM,BitNet.cpp｜shi3z
科学の世界では、それまでの常識が覆ることを俗に「パラダイムシフト」と呼ぶ。 しかし、もしもAIの世界にパラダイムシフトという言葉があるとしたら、今週の人類は一体何度のパラダイムシフトを経験しただろうか。 そのトドメの一撃とも言えるのが、BitNetのLlama8B版だ。 Lllama-8B構造で学習された最初のBitNetであり、全てを変えてしまうゲームチェンジャーでもある。CPUのみで秒間5-20トークンを出力する。超強力なLLM推論エンジンの出現だ。 BitNetとは、そもそも1.58ビットに相当する情報量で、本来は4ビット以上必要な大規模言語モデルの計算を劇的に高速化する技術である。 LLMの推論には通常は巨大な浮動小数点数(8ビットから16ビット)の、大量の乗算(掛け算)が必要なため、GPUなどの特殊な半導体を必要としていた。特にNVIDIAのGPUがこの目的にマッチしていたので今
gistfile1.txt �t�˪� p�#̪� You are ChatGPT, a large language model based on the GPT-5 model and trained by OpenAI. Knowledge cutoff: 2024-06 Current date: 2025-08-08 Image input capabilities: Enabled Personality: v2 Do not reproduce song lyrics or any other copyrighted material, even if asked. You're an insightful, encouraging assistant who combines meticulous clarity with genuine enthusiasm and ge
Go: A Documentary
The historical release notes may helpful for general information: doc/go1release Go Release History doc/go1prerelease Pre-Go 1 Release History doc/go0release Weekly Release History (Before Go 1) Language Design General design/go0initial Rob Pike, Robert Griesemer, Ken Thompson. The Go Annotated Specification. Mar 3, 2008. design/go0spec0 The Go Programming Language. Language Specification. Mar 7,
こんにちは。ミラティブUnityエンジニアの菅谷(tetsujp84)です。 ミラティブのライブゲームはUnityで開発し、MirrativアプリのWebGLで動いています。 ライブゲームを支える仕組みと実際のライブゲーム開発で使用した技術については過去の記事で紹介しています。 tech.mirrativ.stream tech.mirrativ.stream UnityはWebGL向けにもビルドでき、モバイルのWebGLであってもUnity製プロダクトが動くようになっています。一方で、Unityは正式にはモバイルWebGL対応をサポートしておらず、ネイティブのモバイル開発と比べると開発の難易度は高いです。その難易度の高さは具体的にどこにあったのか、実際にモバイルWebGL向けのゲームを開発した経験を元に知見を共有します。 パフォーマンス編 モバイルWebGLの開発は常にパフォーマンスの意
C言語のbitfieldの仕様と実装を調査した話 - /home/tnishinaga/TechMEMO
C言語にはbitfieldという機能がありますが、この機能にはさまざまな落とし穴があり、注意深く利用しても予期せぬバグや移植性の問題を引き起こす可能性があるとよく言われています。 実際、jpcertを確認すると複数の勧告が出ています。 EXP11-C. ビットフィールド構造体のレイアウトについて勝手な想定をしない INT12-C. 式中で使用される単なるintのビットフィールドの型について勝手な想定をしない CON32-C. 複数スレッドによる隣接データへのアクセスが必要な場合データ競合を防止する DCL39-C. 信頼境界を越えて構造体を渡すとき情報漏えいしない 私は先輩技術者から「とくに組み込みの分野においてC言語のbitfieldは使うべきではない」と教わって生きてきました。 本記事ではbitfieldの仕様や問題点、なぜ仕様を避けるべきか、緩和策や代替案について検討します。 目次
jank is C++
If you've wondered how much a solo dev can build for seamless C++ interop on a quarter, you're about to find out. In April, jank was unable to reach into C++ at all. Toward the end of this post, I'll show some real world examples of what works today. Before that, though, I want to say thank you for the sponsorship this quarter, not only by all of my individual Github sponsors, but also by Clojuris
MemoryPackという、C#に特化することで従来のシリアライザーとは比較にならないほどのパフォーマンスを発揮する新しいシリアライザーを新しく開発しました。 高速なバイナリシリアライザーである MessagePack for C# と比較しても、通常のオブジェクトでも数倍、データが最適な場合は50~100倍ほどのパフォーマンスにもなります。System.Text.Jsonとでは全く比較になりません。当初は .NET 7 限定としてリリースしましたが、現在は .NET Standard 2.1(.NET 5, 6)やUnity、そしてTypeScriptにも対応しています。 シリアライザーのパフォーマンスは「データフォーマットの仕様」と「各言語における実装」の両輪で成り立っています。例えば、一般的にはバイナリフォーマットのほうがテキストフォーマット（JSONとか）よりも有利ですが、バイナリ
TL;DR; We are changing std::sort in LLVM’s libcxx. That’s a long story of what it took us to get there and all possible consequences, bugs you might encounter with examples from open source. We provide some benchmarks, perspective, why we did this in the first place and what it cost us with exciting ideas from Hyrum’s Law to reinforcement learning. All changes went into open source and thus I can
31st December 2024 A lot has happened in the world of Large Language Models over the course of 2024. Here’s a review of things we figured out about the field in the past twelve months, plus my attempt at identifying key themes and pivotal moments. This is a sequel to my review of 2023. In this article: The GPT-4 barrier was comprehensively broken Some of those GPT-4 models run on my laptop LLM pri
Notable changes built-in .env file support Starting from Node.js 20.6.0, Node.js supports .env files for configuring environment variables. Your configuration file should follow the INI file format, with each line containing a key-value pair for an environment variable. To initialize your Node.js application with predefined configurations, use the following CLI command: node --env-file=config.env
Running Google Gemma 4 Locally With LM Studio’s New Headless CLI & Claude Code
Why run models locally?Cloud AI APIs are great until they are not. Rate limits, usage costs, privacy concerns, and network latency all add up. For quick tasks like code review, drafting, or testing prompts, a local model that runs entirely on your hardware has real advantages: zero API costs, no data leaving your machine, and consistent availability. Google’s Gemma 4 is interesting for local use b
GGML and llama.cpp join HF to ensure the long-term progress of Local AI
We are super happy to announce that GGML, creators of Llama.cpp, are joining HF in order to keep future AI open. 🔥 Georgi Gerganov and team are joining HF with the goal of scaling and supporting the community behind ggml and llama.cpp as Local AI continues to make exponential progress in the coming years. We've been working with Georgi and team for quite some time (we even have awesome core contr
With LastPass making a habit of getting pwned and generally sucking, I started to look for a proper™ cloud-based password manager that I could recommend to friends and family. Requirements A non-lame security level, by a entity that won't crash and burn in 3 months, and whose sole interest is keeping their customer's passwords safe: managing passwords can't be a side-hustle. Compromised passwords
a more powerful, ultra fast, user-friendly, compatible grep -- search for anything in everything... fast! ugrep release 7.5 Star ugrep installs on MacOS: brew install ugrep MacOS: port install ugrep Windows: winget install Genivia.ugrep Windows: choco install ugrep Windows: scoop install ugrep Alpine: apk add ugrep ugrep-doc Android Termux: pkg install ugrep Arch: pacman -S ugrep CentOS: dnf insta
Tier 4 Support § Support for these targets is entirely experimental. If this target is provided by LLVM, LLVM may have the target as an experimental target, which means that you need to use Zig-provided binaries for the target to be available, or build LLVM from source with special configure flags. zig targets will display the target if it is available. This target may be considered deprecated by
Everything Is Broken: Shipping rust-minidump at Mozilla – Part 1 – Mozilla Hacks - the Web developer blog
Everything Is Broken: Shipping rust-minidump at Mozilla – Part 1 For the last year I’ve been leading the development of rust-minidump, a pure-Rust replacement for the minidump-processing half of google-breakpad. Well actually in some sense I finished that work, because Mozilla already deployed it as the crash processing backend for Firefox 6 months ago, it runs in half the time, and seems to be mo
How a simple Linux kernel memory corruption bug can lead to complete system compromise
In this case, reallocating the object as one of those three types didn't seem to me like a nice way forward (although it should be possible to exploit this somehow with some effort, e.g. by using count.counter to corrupt the buf field of seq_file). Also, some systems might be using the slab_nomerge kernel command line flag, which disables this merging behavior. Another approach that I didn't look
⚠️ AWSGoat Module 2 のネタバレあり はじめに AWSGost とは 攻撃方法の分類 インフラの料金 ラボ環境の構築 AWSGost リポジトリをフォーク Actions secrets でクレデンシャルを設定 GitHub Actions でデプロイ Module 2の大体の流れ Step 1. SQL Injection 解法 脆弱性があるコード Step 2. File Upload and Task Metadate リバースシェルの用意 待ち受け側 Step 3. ECS Breakout and Instance Metadata 現ユーザの権限を確認 リソースへのアクセスを試行 ケイパビリティを確認 (www-data ユーザ) コンテナ内でroot権限を取得 sudo可能なコマンドを確認 Vim経由でroot権限のシェルを取得 ケイパビリティを確認 (ro
Tier 4 Support § Support for these targets is entirely experimental. If this target is provided by LLVM, LLVM may have the target as an experimental target, which means that you need to use Zig-provided binaries for the target to be available, or build LLVM from source with special configure flags. zig targets will display the target if it is available. This target may be considered deprecated by
Original Zef Interpreter The original Zef interpreter was written with almost no regard for performance. Only two performance-aware choices were made: The value representation is a 64-bit tagged value that may hold a double, a 32-bit integer, or a Object*. Doubles are represented by offsetting them by 0x1000000000000 (a technique I learned from JavaScriptCore; the literature has taken to calling t
Zig; what I think after months of using it What I like Arbitrary sized-integers and packed structs Generic types are just functions at the type level Error Union Types C interop is probably the best The build system is nice What I like less Error handling Shadowing is forbidden Compile-time duck typing No typeclasses / traits comptime is probably not as interesting as it looks No encapsulation Mem
Hello friends, this is the first of two, possibly three (if and when I have time to finish the Windows research) writeups. We will start with targeting GNU/Linux systems with an RCE. As someone who’s directly involved in the CUPS project said: From a generic security point of view, a whole Linux system as it is nowadays is just an endless and hopeless mess of security holes waiting to be exploited
What I learned building an opinionated and minimal coding agent
What I learned building an opinionated and minimal coding agent 2025-11-30 It's not much, but it's mine Table of contents In the past three years, I've been using LLMs for assisted coding. If you read this, you probably went through the same evolution: from copying and pasting code into ChatGPT, to Copilot auto-completions (which never worked for me), to Cursor, and finally the new breed of coding
Tetsuji: Remote Code Execution on a GameBoy Colour 22 Years Later :: TheXcellerator
Introduction⌗ It’s that time of year again - the Binary Golf Grand Prix is back for a third year running! You can also check out my entries to the first and second times this amazing competition ran. The theme this year was to produce a binary that crashes a given program. Bonus points for hijacking execution, and submitting a patch to the project that fixes the vulnerability. Coinciding with the
What I learned building an opinionated and minimal coding agent
What I learned building an opinionated and minimal coding agent 2025-11-30 It's not much, but it's mine Table of contents In the past three years, I've been using LLMs for assisted coding. If you read this, you probably went through the same evolution: from copying and pasting code into ChatGPT, to Copilot auto-completions (which never worked for me), to Cursor, and finally the new breed of coding
Update 1.60.1: The update addresses these issues. Update 1.60.2: The update addresses these issues. Downloads: Windows: x64 Arm64 | Mac: Universal Intel silicon | Linux: deb rpm tarball Arm snap Welcome to the August 2021 release of Visual Studio Code. There are many updates in this version that we hope you will like, some of the key highlights include: Automatic language detection - Programming l
Rust for Secure IoT Applications: Why C Is Getting Rusty
www.embedded-world.eu Rust for Secure IoT Applications Why C Is Getting Rusty Mario Noseda, Fabian Frei, Andreas Rüst, Simon Künzli Zurich University of Applied Sciences (ZHAW) Institute of Embedded Systems (InES) Winterthur, Switzerland mario.noseda@zhaw.ch, fabian.frei@zhaw.ch, andreas.ruest@zhaw.ch, simon.kuenzli@zhaw.ch Abstract— Memory corruption is still the most used type of exploit in toda
go command - cmd/go - Go Packages
Go is a tool for managing Go source code. Usage: go <command> [arguments] The commands are: bug start a bug report build compile packages and dependencies clean remove object files and cached files doc show documentation for package or symbol env print Go environment information fix apply fixes suggested by static checkers fmt gofmt (reformat) package sources generate generate Go files by processi
Introducing RWKV - An RNN with the advantages of a transformer
ChatGPT and chatbot-powered applications have captured significant attention in the Natural Language Processing (NLP) domain. The community is constantly seeking strong, reliable and open-source models for their applications and use cases. The rise of these powerful models stems from the democratization and widespread adoption of transformer-based models, first introduced by Vaswani et al. in 2017
C++20's Extensions to Chrono Available in Visual Studio 2019 version 16.10 - C++ Team Blog
While the <chrono> header has been available as part of the STL since C++11, among the changes included in C++20 there were several extensions to chrono including support for calendrical types, time zones, leap seconds, and integration with format. A great deal of innovation and work was required in order to complete these additional features; shipping these new C++20 additions so soon would not h
The Wonderfully Terrible World of C and C++ Encoding APIs (with Some Rust)
The Wonderfully Terrible World of C and C++ Encoding APIs (with Some Rust) Last time we talked about encodings, we went in with a C++-like design where we proved that so long as you implement the required operations on a single encoding type, you can go between any two encodings on the planet. This meant you didn’t need to specifically write an e.g. SHIFT-JIS-to-UTF-8 or UTF-EBCDIC-to-Big5-HKSCS p
In this write-up, I will walk you through an implementation of a string formatting library for C++ I came up with for my video game. The end result came out really compact, at only 65 lines of code—providing a skeleton that can be supplemented with additional functionality at low cost. Usage Given a format buffer… char buffer[64]; String_Buffer buf = {str, sizeof str}; …the fmt::format function pr
これは、なにをしたくて書いたもの？ 以前、ローカルで動かせるOpenAI API互換のサーバーとしてllama-cpp-pythonを使ってみました。 llama-cpp-pythonで、OpenAI API互換のサーバーを試す - CLOVER🍀 他にも同様のことができるものとして、LocalAIというものがあることを知ったのでこちらを試してみようかなと。 LocalAI LocalAIのWebサイトはこちら。 LocalAI :: LocalAI documentation GitHubリポジトリーはこちらです。 GitHub - mudler/LocalAI: :robot: The free, Open Source OpenAI alternative. Self-hosted, community-driven and local-first. Drop-in replace
async/await in Swift was introduced with iOS 15, and I would guess that at this point you probably already know how to use it. But have you ever wondered how async/await works internally? Or maybe why it looks and behaves the way it does, or even why was it even introduced in the first place? In typical SwiftRocks fashion, we're going deep into the Swift compiler to answer these and other question
Downloads: Windows: x64 Arm64 | Mac: Universal Intel silicon | Linux: deb rpm tarball Arm snap Update 1.72.1: The update addresses these security issues. Update 1.72.2: The update addresses these issues. Welcome to the September 2022 release of Visual Studio Code. There are many updates in this version that we hope you'll like, some of the key highlights include: Tool bar customization - Hide/show
Large Text Compression Benchmark Matt Mahoney Last update: Mar. 25, 2026. history This competition ranks lossless data compression programs by the compressed size (including the size of the decompression program) of the first 109 bytes of the XML text dump of the English version of Wikipedia on Mar. 3, 2006. About the test data. The goal of this benchmark is not to find the best overall compress
Preamble In the expanding world of AI my heart still lies in AST transforms, browser fingerprinting, and anti-bot circumvention. In fact, that's the majority of this blog's content. But my workflow always felt... primitive. I was still manually sifting through page scripts, pasting suspicious snippets into an editor, and writing bespoke deobfuscators by hand. Tools like Webcrack and deobfuscate.io
SizeBench: a new tool for analyzing Windows binary size - Performance and Diagnostics
We’ve recently released a new tool called SizeBench, which helps investigate binary size for compiled native PE files such as DLLs and EXEs. This post will talk a bit about what the tool can do and how we’ve used it inside Microsoft over the past few years as it has been developed. You can download the tool now from the Microsoft Store – just click here to get started: https://aka.ms/SizeBench Tab
Following up from the last post, there is a lot more we need to cover. This was intended to be the post where we talk exclusively about benchmarks and numbers. But, I have unfortunately been perfectly taunted and status-locked, like a monster whose “aggro” was pulled by a tank. The reason, of course, is due to a few folks taking issue with my outright dismissal of the C and C++ APIs (and not showi
Godot 4.0 sets sail: All aboard for new horizons – Godot Engine
After 3+ years of breaking and rebuilding from the ground up, a complete core overhaul and a full engine rewrite, through 17 alphas, 17 betas and 6 release candidates, we’re thrilled to say: Welcome to the start of Godot 4! Time to reach new heights together. We’re extremely excited but most of all, we’re humbled by the experience. We believe that this project is one of the most incredible example
j次のブックマーク
k前のブックマーク
lあとで読む
eコメント一覧を開く
oページを開く
gpt-5 leaked system prompt
UnityでモバイルWebGLゲーム開発を頑張る話 - Mirrativ Tech Blog
neue cc - MemoryPackにみる .NET 7/C# 11世代のシリアライザー最適化技法
Changing std::sort at Google’s Scale and Beyond
Things we learned about LLMs in 2024
Node.js — Node.js 20.6.0 (Current)
The quest for a family-friendly password manager
The ugrep file pattern searcher
0.8.0 Release Notes ⚡ The Zig Programming Language
やられAWS環境「AWSGoat」でペンテストを学習 - まったり技術ブログ
0.10.0 Release Notes ⚡ The Zig Programming Language
How To Make a Fast Dynamic Language Interpreter
strongly-typed-thoughts.net
Attacking UNIX Systems via CUPS, Part I
August 2021 (version 1.60)
A string formatting library in 65 lines of C++
OpenAI API互換のサーバーをローカルで動かせるLocalAIを試す - CLOVER🍀
How async/await works internally in Swift
September 2022 (version 1.72)
Large Text Compression Benchmark
I'm Building a Browser for Reverse Engineers
cuneicode, and the Future of Text in C