はじめにこの記事では効率的な合成データ生成からそのデータを学習したモデルのGGUF変換、OllamaやLM Studioでの推論まで行います。 データ合成にはSDG LOOM、学習にはUnsloth Studio、推論にはLM Studioを用います。 これを理解すれば誰でもオリジナルのLLMを作成することができます。 今回は「小説生成ローカルモデル」を例に挙げて作成を行います。 それでは初めて行きましょう。 合成データの作り方このステップでは、LLMを用いた合成データを作ります。 オリジナルのLLMを作成するにあたって、1番大事なのは合成データです。 どのようなデータを作り、そしてそれを用いてどのような大規模言語モデルを目指すのかを決定するのがこのフェーズになります。 合成データを作るにあたって、以下の3つの点が大切だと個人的に思っています。 ・データの品質 ・コストパフォーマンス そし
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
Internet Engineering Task Force (IETF) K. Davis Request for Comments: 9562 Cisco Systems Obsoletes: 4122 B. Peabody Category: Standards Track Uncloud ISSN: 2070-1721 P. Leach University of Washington May 2024 Universally Unique IDentifiers (UUIDs) Abstract This specification defines UUIDs (Universally Unique IDentifiers) -- also known as GUIDs (Globally Unique IDentifiers) -- and a Uniform Resou
Over the course of my Spring 2020 semester at Harvey Mudd College, I developed a self-hosting compiler entirely from scratch. This article walks through many interesting parts of the project. It’s laid out so you can just read from beginning to end, but if you’re more interested in a particular topic, feel free to jump there. Or, take a look at the project on GitHub. Table of contents What the pro
Patterns for Building LLM-based Systems & Products [ llm engineering production 🔥 ] · 66 min read Discussions on HackerNews, Twitter, and LinkedIn “There is a large class of problems that are easy to imagine and build demos for, but extremely hard to make products out of. For example, self-driving: It’s easy to demo a car self-driving around a block, but making it into a product takes a decade.”
2025-05-06, Version 24.0.0 (Current), @RafaelGSS and @juanarbol We’re excited to announce the release of Node.js 24! This release brings several significant updates, including the upgrade of the V8 JavaScript engine to version 13.6 and npm to version 11. Starting with Node.js 24, support for MSVC has been removed, and ClangCL is now required to compile Node.js on Windows. The AsyncLocalStorage API
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
Shai Hulud Strikes Again (v2)Another wave of Shai-Hulud campaign has hit npm with more than 500 packages and 700+ versions affected. Update: November 26, 2025 PostHog has published a detailed post mortem describing how one of its GitHub Actions workflows was abused as an initial access vector for Shai Hulud v2. An attacker briefly opened a pull request that modified a script executed via pull_requ
Transformer models: an introduction and catalog — 2023 Edition January 16, 2023 52 minute read This post is now an ArXiV paper that you can print and cite. Update 05/2023 Another pretty large update after 4 months. I was invited to submit the article to a journal, so I decided to enlist some help from some LinkedIn colleages and completely revamp it. First off, we added a whole lot of new models,
はじめに：パーソナライズされた英語リスニング教材の必要性 英語学習、特にリスニング力の向上において、自身の興味やレベルに合致した教材を見つけることは重要な要素です。しかし、市販の教材やオンラインサービスでは、特定のトピックや細かなレベル調整に対応しきれないケースが多くありました。 私自身、特定の専門分野に関する議論を、自身の英語レベルに適した難易度で聞きたいと考えていましたが、既存の教材ではなかなか見つけることができませんでした。 この課題を解決するため、「自分の興味・レベルに合わせて、リスニング教材を生成できるツール」の必要性を感じ、開発に至りました。本記事では、その開発経緯、技術的な実装について共有します。 デモ動画 課題設定：理想的な英語リスニング教材の要件 開発にあたり、理想的なリスニング教材に必要な要件を以下のように定義しました。 トピックの自由度: ユーザーが任意のキーワードを
Introduction: Density Is King (With a Tiny VM) I've previously come to the conclusion that there's little reason for using bytecode in the modern world, except in order to get more compact code, for which it can be very effective. So, what kind of a bytecode engine will give you more compact code? Suppose I want a bytecode interpreter for a very small programming environment, specifically to minim
An interesting twist in my recent usage of SQLite was the fact that I noticed my research scripts and the database intertwine more. SQLite is unique in that it really lives in-process, unlike standalone database servers. There is a feature to that which does not get used very frequently, but can be indispensable in some situations. By the way, the talk about the system that made me me to explore S
C interpreters are a common language implementation technique and the basis for the reference implementations of languages such as Lua, Ruby, and Python. Unfortunately, C interpreters are slow, especially compared to language implementations powered by JIT compilers. In this post I’m going to show that it is possible to take C interpreters and, by changing a tiny proportion of code, automatically
Noble Numbat Release Notes Table of Contents Introduction New features in 24.04 LTS Known Issues Official flavours More information Introduction These release notes for Ubuntu 24.04 LTS (Noble Numbat) provide an overview of the release and document the known issues with Ubuntu and its flavours. For details of the changes applied since 24.04, please see the 24.04.2 change summary. Support lifespan
Tetris That Hates You StickManStickMan #611, by Sam Hughes. HATETRIS is a version of Tetris written in 2010 by programmer and sci-fi author Sam Hughes. According to his initial description of the game: This is bad Tetris. It’s hateful Tetris. It’s Tetris according to the evil AI from “I Have No Mouth And I Must Scream”. (And if you aren’t familiar with Tetris at all, and don’t know the rules or pi
The sad state of property-based testing libraries Posted on Jul 2, 2024 Property-based testing is a rare example of academic research that has made it to the mainstream in less than 30 years. Under the slogan “don’t write tests, generate them” property-based testing has gained support from a diverse group of programming language communities. In fact, the Wikipedia page of the original property-bas
I find blockchain fascinating because it extends open source software development to open source + state. This seems to be a genuine/exciting innovation in computing paradigms; We don’t just get to share code, we get to share a running computer, and anyone anywhere can use it in an open and permissionless manner. The seeds of this revolution arguably began with Bitcoin, so I became curious to dril
Why APL is a language worth knowing
“A language that doesn't affect the way you think about programming, is not worth knowing.”, by Alan J. Perlis. Why APL is a language worth knowing Alan Perlis, the computer scientist recipient of the first Turing award, wrote “A language that doesn't affect the way you think about programming, is not worth knowing.” ― Alan J. Perlis, 1982. Special feature: Epigrams on programming. ACM Sigplan Not
How I hacked SONOS and YouTube the same day Or how to play YouTube videos on you SONOS, easy and for free. SONOS is a brand of connected speakers that allow users on the network listen to music that they can choose using the SONOS specific Android / iOS application, or 3d party services with monthly subscription like Deezer / Spotify / YouTube Music from their local network. SONOS has been growing
0. はじめに LayerX Ai Workforce事業部でR&Dチームマネージャーの澁井（しぶい）と申します。 実業務でLLMやAIエージェントを活用するときに頻繁に課題になることとして、作ったLLM/AIエージェントシステムを評価するデータが足りない、ということがあります。こうした課題に対処するため、LLMやAIエージェントを用いて合成データを作ることは一般的なプラクティスと言えます。しかし、必要な品質の合成データを大量かつ多様に作ることは相応に難しく、エンジニアリングが伴います。 本テックブログでは、合成データの作り方に関するTips集を紹介します。このTipsが読者の合成データ作成に貢献できると幸いです。 1. 合成データとは何か？ AIエージェント時代に注目される理由 合成データ（Synthetic Data）とは、実データの統計的特性や意味構造を保ちながら、プログラムやモデル
Codified Context: Infrastructure for AI Agents in a Complex Codebase Aristidis Vasilopoulos Independent Researcher USA arisvas4@gmail.com Abstract LLM-based agentic coding assistants lack persistent memory: they lose coherence across sessions, forget project conventions, and repeat known mistakes. Recent studies characterize how develop- ers configure agents through manifest files, but an open cha
Playwright Agents を Codex CLI に対応させブラウザ自動テストを実装してもらう（公式が未サポートだけどできました） - GMOインターネットグループ グループ研究開発本部
2025.11.05 Playwright Agents を Codex CLI に対応させブラウザ自動テストを実装してもらう（公式が未サポートだけどできました） AI ツール大好き D.M.です。 結論ファースト ・Playwright Agents は LLM と Playwright MCP でブラウザ自動テストを設計、実装、修正ができるAIエージェント。 ・Playwright Agents は Codex CLI 未対応だが、 AGENTS.md をフォルダを分けて作成することでほぼ問題なく利用できる。 このブログでやりたいこと Codex CLI で Playwright Agents を呼び出し、ブラウザ自動テストを設計実装してもらいたい。 感触としては結構簡単に実現できます。 [アジェンダ] 導入編： Playwright Agents を Codex CLI で動かす →
Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models Qizheng Zhang 1∗ Changran Hu 2∗ Shubhangi Upasani 2 Boyuan Ma 2 Fenglu Hong 2 Vamsidhar Kamanuru 2 Jay Rainton 2 Chen Wu 2 Mengmeng Ji 2 Hanchen Li 3 Urmish Thakker 2 James Zou 1 Kunle Olukotun 1 1 Stanford University 2 SambaNova Systems, Inc. 3 UC Berkeley ∗ equal contribution # qizhengz@stanford.edu, changran.hu@sa
GitHub - ComfyUI-Workflow/awesome-comfyui: A collection of awesome custom nodes for ComfyUI
ComfyUI-Gemini_Flash_2.0_Exp (⭐+172): A ComfyUI custom node that integrates Google's Gemini Flash 2.0 Experimental model, enabling multimodal analysis of text, images, video frames, and audio directly within ComfyUI workflows. ComfyUI-ACE_Plus (⭐+115): Custom nodes for various visual generation and editing tasks using ACE_Plus FFT Model. ComfyUI-Manager (⭐+113): ComfyUI-Manager itself is also a cu
[Simon Tatham, initial version 2023-09-01, last updated 2025-03-25] [Coroutines trilogy: C preprocessor | C++20 native | general philosophy ] Introduction Why I’m so enthusiastic about coroutines The objective view: what makes them useful? Versus explicit state machines Versus conventional threads The subjective view: why do I like them so much? “Teach the student when the student is ready” They s
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
一日でできる！ オリジナルのローカルLLMの作り方【データ合成からLM Studioまで】｜Holy_fox
Changing std::sort at Google’s Scale and Beyond
RFC 9562: Universally Unique IDentifiers (UUIDs)
Kalyn: a self-hosting compiler for x86-64
Patterns for Building LLM-based Systems & Products
Node.js — Node.js 24.0.0 (Current)
0.10.0 Release Notes ⚡ The Zig Programming Language
Shai Hulud Strikes Again (v2) - Socket
Transformer models: an introduction and catalog — 2023 Edition
理想の英語アプリが無かったので作った - OpenAI TTS API, Gradio🚀
bytecode interpreters for tiny computers ⁑ Dercuano
Supercharge SQLite with Ruby functions
Laurence Tratt: Retrofitting JIT Compilers into C Interpreters
Ubuntu 24.04 LTS (Noble Numbat) Release Notes
Getting the World Record in HATETRIS
The sad state of property-based testing libraries
A from-scratch tour of Bitcoin in Python
How I hacked SONOS and YouTube the same day
AIエージェントを強くする『合成データ』作成のニッチなTips集 - LayerX エンジニアブログ
Codified Context: Infrastructure for AI Agents in a Complex Codebase
Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models
Philosophy of coroutines