LCM LoRA の概要｜npaka
以下の記事が面白かったので、かるくまとめました。 ・SDXL in 4 steps with Latent Consistency LoRAs 1. はじめに「LCM」 (Latent Consistency Model) は、元モデルを別モデルに蒸留することで、画像生成に必要なステップ数を減らす手法です。25～50ステップかかっていた処理を4～8ステップで可能にします。 蒸留したモデルは、より小さくなるように設計される場合 (DistilBERT、Distil-Whisperなど)と、必要なステップ数が少なくなるように設計される場合があります。これは通常、膨大な量のデータ、忍耐力、少数のGPU を必要とし、時間とコストが必要でした。しかし、それも過去の話です。 11月9日 、「LCM」で蒸留したかのように「Stable Diffusion」「SDXL」を本質的に高速化できる新手法「LCM
An oral history of Bank Python November 2021 The strange world of Python, as used by big investment banks High finance is a foreign country; they do things differently there Today will I take you through the keyhole to look at a group of software systems not well known to the public, which I call "Bank Python". Bank Python implementations are effectively proprietary forks of the entire Python ecos
Now that the name is official, we've got more details about Sony's next-gen console—from the haptics-packed controller to UI improvements. Sony's PlayStation 5 might just meet this generation's need for immediacy.Photograph: Chris Ratcliffe/Getty Images Ever since the original PlayStation hit the market in 1994, Sony's series of videogame consoles has stuck to the numbers. No "Super," no "Max," no
Celebrate 15 Years of Amazon S3 with ‘Pi Week’ Livestream Events | Amazon Web Services
AWS News Blog Celebrate 15 Years of Amazon S3 with ‘Pi Week’ Livestream Events I wrote the blog post that announced Amazon Simple Storage Service (Amazon S3) fifteen years ago today. In that post, I made it clear that the service was accessed via APIs and that it was targeted at developers, outlined a few key features, and shared pricing information. Developers found that post, started to write co
Hardik Pandya
Improving your sense of judgment is critical to becoming better at designing products. Good design principles can help improve your judgment. When you’re faced with conflicting choices in the design process, good principles help you break the tie. Good principles go beyond common truisms like ‘Our designs should be accessible’, ‘We should delight the user’ and so on. Good principles uncover the of
Let me introduce competitive programming - GMO Research & AI Tech Blog
Self-Introduction ※今回は英文で執筆いただきました。 ▼和文バージョンはこちら。 Hello ! I am Priyanka, currently working as System Engineer in GMOリサーチ株式会社. It has been almost 3 years since I’ve started working and writing codes for real products in this company. You can say that I am still in my learning phase as an engineer. I like to achieve new things and always have a mindset to continue growing, so I find ways to improve
Posted by Stephanie Cuthbertson, Senior Director of Product Management, Android After more than a year of development and months of testing by early adopters, we’re ready to introduce Android 10 to the world! Android 10 is built around three important themes. First, Android 10 is shaping the leading edge of mobile innovation with advanced machine-learning and support for emerging devices like fold
GitHub - diff-usion/Awesome-Diffusion-Models: A collection of resources and papers on Diffusion Models
A Survey on Video Diffusion Models Zhen Xing, Qijun Feng, Haoran Chen, Qi Dai, Han Hu, Hang Xu, Zuxuan Wu and Yu-Gang Jiang arXiv 2023. [Paper] 16 Oct 2023 State of the Art on Diffusion Models for Visual Computing Ryan Po, Wang Yifan, Vladislav Golyanik, Kfir Aberman, Jonathan T. Barron, Amit H. Bermano, Eric Ryan Chan, Tali Dekel, Aleksander Holynski, Angjoo Kanazawa, C. Karen Liu, Lingjie Liu, B
Amazon Time Sync Service now makes it easier to generate and compare timestamps
Amazon Time Sync Service now allows you to easily generate and compare timestamps from Amazon EC2 instances with ClockBound, an open source daemon and library. This information is valuable to determine order and consistency for events and transactions across EC2 instances, independent from the instances’ respective geographic locations. ClockBound calculates your Amazon EC2 instance’s clock error
This is our goal architecture design, please read the article to understand the journey :)This blog article is participating in the Mercari Bold Challenge month (#6) Hi everyone, this is Raphael from the Microservices Platform team at Mercari. Bluntly introduced, we are a post-IPO Japanese C2C (Customer to Customer) marketplace transitioning from a monolithic to a microservices architecture. A few
MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model
TL;DR: We propose MagicAnimate, a diffusion-based human image animation framework that aims at enhancing temporal consistency, preserving reference image faithfully, and improving animation fidelity.
Solving Quantitative Reasoning Problems With Language Models
Solving Quantitative Reasoning Problems with Language Models Aitor Lewkowycz∗, Anders Andreassen†, David Dohan†, Ethan Dyer†, Henryk Michalewski†, Vinay Ramasesh†, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, Yuhuai Wu, Behnam Neyshabur∗, Guy Gur-Ari∗, and Vedant Misra∗ Google Research Abstract Language models have achieved remarkable performance on a wide range of tasks that require
The 2019 Design Systems Survey by Sparkbox
The 2019 Design Systems Survey is intended to illuminate how design systems are created, used, and maintained. Design systems have the ability to solve real-life problems by establishing consistency in design and code, promoting efficiency when creating products, communicating usage guidance, encouraging reuse instead of rework, increasing the accessibility and usability of digital properties, and
Introducing Docker Build Cloud: A new solution to speed up build times and improve developer productivity | Docker
Introducing Docker Build Cloud: A New Solution to Speed Up Build Times and Improve Developer Productivity Developers face a growing predicament — the long wait times for builds to complete. In fact, the average build time increased by an average of 15.9% between 2020 and 2021, according to a survey by Incredibuild. On average, developers lose around one hour each day, the study says, and this dela
※この記事はC++20を相談しながら調べる会 #3の成果として書かれました。 P1152R4 : Deprecating volatileを読み解く過程に生じたvolatileについての調査（脱線）メモです。ほぼ文章です。 C++におけるvolatileの効果 マルチスレッドとvolatile 用途と副次的効果 アウトオブオーダー実行とvolatile 参考文献 謝辞 C++におけるvolatileの効果 C++におけるvolatile指定の効果は次のように規定されています（6.9.1 Sequential execution [intro.execution]）。 Reading an object designated by a volatile glvalue ([basic.lval]), modifying an object, calling a library I/O fun
GitHub - magmide/magmide: A dependently-typed proof language intended to make provably correct bare metal code possible for working software engineers.
This repo is still very early and rough, it's mostly just notes, speculative writing, and exploratory theorem proving. Most of the files in this repo are just "mad scribblings" that I haven't refined enough to actually stand by! If you prefer video, this presentation talks about the core ideas that make formal verification and Magmide possible, and the design goals and intentions of the project: I
DGS Framework
Getting Started Introduction¶ The DGS framework makes it easy to create GraphQL services with Spring Boot. The framework provides an easy-to-use annotation based programming model, and all the advanced features needed to build and run GraphQL services at scale. The DGS framework is primarily maintained by Netflix, surrounded by an active community. At Netflix we build our GraphQL architecture on t
年商150億のビジネスオーナーが出版の依頼を断った理由 - Evergreen Blog ～永続的にお金とファンを創造する資産ブログの作り方～
今日は短めの記事になりますが、 ブログを書く人にとって重要な話なのでシェアしたいと思います。 この話は年商150億円のビジネスオーナーの人から聞いた話です。 その方のところに出版社から本の執筆依頼が来たそうなのですが、 その方は出版依頼を断ったそうです。 その理由は自分の伝えたいことではなく、 出版社の指示したテーマについて本を書いてほしいと言われたからです。 そのテーマとは「がんばらなくてもいいんだよ」的な内容だったそうです。 良い本と売れる本は違う 情報発信＝キャラクタービジネス 主張を一貫させる まとめ 良い本と売れる本は違う そのビジネスオーナーの方はこう言っていました。 「良い本とベストセラーになる本は違う」と。 一般的に良い本であれば売れるというイメージがありますが、 売れている本だからといってそれが良書であるとは限りません。 売れる本とはいったいどういう本だと思いますか？ そ
I’m excited to announce Swift Algorithms, a new open-source package of sequence and collection algorithms, along with their related types. Algorithms are powerful tools for thought because they encapsulate difficult-to-read and error-prone raw loops. The Algorithms package includes a host of powerful, generic algorithms frequently found in other popular programming languages. We hope this new pack
How do you manage your Machine Learning Experiments?
‘Every experiment is sacred Every experiment is great If an experiment is wasted God gets quite irate’ ~Sacred Here I come clean, for a long time I have been a caveman. I have been using spreadsheets to log my ML experiments, all started well, I was happy, and then a deadline came and all of a sudden it went messy, very messy... I trusted my self-discipline to keep consistency and it failed me. I
週刊Railsウォッチ（20200616後編）本番環境をFullstaq Rubyに換えた理由、CSRF発生フローチャート、DBのトランザクション分離レベル比較ほか｜TechRacho by BPS株式会社
morimorihoge注：本トピックに関するコメントとして、初出時人種差による不平等を容認・揶揄するようにも取れる内容がありましたので当該部分を削除させていただきました。 Twitter上でのご指摘を受けて社内ヒアリングを行ったところ、私を含む参加メンバーで主題の認識違いがあり、主題のすれ違いの中話していた内容をテキスト化した際に公開記事として掲載するにあたって不適切な内容となってしまっていたことが判明いたしました。不快に感じられた方にはこの場を借りてお詫びいたします。 ※BPS株式会社は会社・組織として人種その他先天的・後天的な要因による差別を容認・支持する立場にはありません（弊社代表にも確認済み） 各記事冒頭には⚓でパーマリンクを置いてあります: 社内やTwitterでの議論などにどうぞ 「つっつきボイス」はRailsウォッチ公開前ドラフトを（鍋のように）社内有志でつっついたときの会
Structuring your Storybook
The more information you have, the harder it is to find what you're looking for. At first, your Storybook has a handful of components, so it's easy to keep track of things. But as the number of components grows, so does the organizational complexity. You end up with questions like “Where did we put that component again?” and “Is there a story for this state?” That leads to confusion, or worse, dup
データ分析をおこなう人のほとんどは「一般化可能性」とか「外的妥当性」という言葉を耳にしたことがあると思います。 例えば次のような議論を聞いたことがないでしょうか。 ・「アメリカ人のデータから●●を食べると死ににくいことがわかった」→「いやいや、日本人のデータではないから日本人が●●を食べて健康になれるかどうかはわからない」 ・「大学生にボランティアで実験に参加してもらって▲▲が判明した」→「大学生以外には当てはまらんでしょう？」 ・「SNSでアンケート調査をして、AとBの関連を検討した」→「SNSを使っているひと、さらに調査に協力してくれる人、って偏りすぎて参考にならない」 論文を書く際にも、最後の方で研究の限界点を述べる時に ・「本研究は●●というサンプルを使っているので一般化可能性は限定的である」 ・「小規模RCTなので実在集団への一般化はできないかもしれない。より広い集団でさらなる研
Why Authorization is Hard
Feb 2023 Update: Since writing this post in 2021, we've built, released, and GA-ed Oso Cloud: our opinionated solution for authorization. Two years ago, my cofounder and I started building security tools for infrastructure. We kept hearing that application developers were building their own homegrown authorization tools. At first we were a little skeptical. People have been building authorization
ControlNet image sequence 2 image sequence script v2 Added TemporalNet controlnet model support. Using this, it is possible to increase consistency between image sequences by inputting images from previous frames. Download the model file from here and put it in the same location as your other controlnet models. diff_control_sd15_temporalnet_fp16.safetensors · CiaraRowles/TemporalNet at main (huggi
Font size is useless; let’s fix it Translations: Chinese Japanese Russian What happens when you set "font_size": 32 in your favorite editor? I would’ve told you anyway, but I’m glad that you asked. Let’s try to guess. I am using Sublime Text 4 on macOS: If we measure letters themselves, number 32 is nowhere to be seen: 32 is not the width or height of a letter, neither is it a capital letter heigh
Dendron
For full text search please use the '?' prefix. e.g. ? Onboarding Dendron Dendron is an open-source, local-first, markdown-based, note-taking tool. Think of it as a cache for everything that you care about - if you've spent more then five minutes solving a problem, you should never spent any more time solving the same exact problem. Dendron is a knowledge base built by and for developers and integ
Want help enforcing rules? 👮 Gently nudge you and your team into ECSS. Install the Stylelint config Eager to build something? 🏁 A quick & efficient starting point is within reach! Clone the scaffolding library (WIP) Want to know the beast? 📰 Go ahead and see what you think of it! Dive into ECSS rules Curious about it all? Well, keep scrolling! Why ECSS? Because CSS as a language is misunderstoo
What’s New In Python 3.9
What’s New In Python 3.9¶ Editor: Łukasz Langa This article explains the new features in Python 3.9, compared to 3.8. Python 3.9 was released on October 5, 2020. For full details, see the changelog. Summary – Release highlights¶ New syntax features: PEP 584, union operators added to dict; PEP 585, type hinting generics in standard collections; PEP 614, relaxed grammar restrictions on decorators. N
What makes a great technical blog | notes.eatonphil.com
I want to explain why the blogs in My favorite technical blogs are my favorite. That page is solely about non-corporate tech blogs. So this post is too. I'll have to make another list for favorite corporate tech blogs. In short, they: Tackle hard and confusing topics Show working code Make things simpler Write regularly Talk about tradeoffs and downsides Avoid internet slang, memes, swearing, sarc
ML Education at Uber: Frameworks Inspired by Engineering Principles
You’re seeing information for Poland . To see local features and services for another location, select a different city. Show more Introduction At Uber, millions of machine learning (ML) predictions are made every second, and hundreds of applied scientists, engineers, product managers, and researchers work on ML solutions daily. Uber wins by scaling machine learning. We recognize org-wide that a p
Buffer | State Of Remote Work 2020
When some people think of the workplace of the future, they envision futuristic-style holograms having a meeting or robots cooking lunch for everyone in the office. Increasingly, though, the workplace of the future is looking more simple — people having the flexibility to work remotely from home with teammates all around the world. With that in mind, the question is no longer “is remote work here
今日の話題はこちら 最近話題のリアルタイムAIお絵描きを実際にやってみようという回です。 使用技術 今回はタイトル通りDiffusersとLCM-Loraを使います。 AIによって画像生成をリアルタイムで行いながら、お絵描き補助をするというツールは実は画像生成AIが出てきた初期の方にあったりするのですが、今回はLCMという技術を使うことで画像生成速度が大幅に向上したことで、ほぼリアルタイムに書いたものがAIによって高品質化されるという点が相違点になります。 LCM及びLCM-LoRAとは 今回肝となるLCMですが、正式名称はlatent-consistency-modelといいます。 LCMが出てきたのは2023/10/6に公表された画像生成の高速化技術で、ものすごく簡潔に説明すると、従来の画像生成AIが複数のステップを踏みながらノイズを徐々に除去して画像生成するところを、一発でノイズ画像
オミータです。ツイッターで人工知能のことや他媒体で書いている記事など を紹介していますので、人工知能のことをもっと知りたい方などは気軽に@omiita_atiimoをフォローしてください！ 他にも次のような記事を書いていますので興味があればぜひ！ GANにもオーギュメンテーションは超有効だった！DAだけでSoTA達成した最新論文を解説！ 画像認識の定番データセットImageNetはもう終わりか パラメータ数10億！最新の巨大画像認識モデル「BiT」爆誕 & 解説 【基本編】画像認識に使用されるData Augmentationを一挙にまとめてみた！ 画像認識の最新SoTAモデル「Noisy Student」を徹底解説！ 画像生成分野で物凄い成果を出し続けているモデルとしてGenerative Adversarial Networks、通称GANがあります。GANは基本的に 「生成器」と「識
IntroductionInspired by the cross-functional cooperation promised by DevOps, platform engineering has begun to emerge in enterprises as an explicit form of that cooperation. Platforms curate and present foundational capabilities, frameworks and experiences to facilitate and accelerate the work of internal customers such as application developers, data scientists and information workers. Particular
Building top-tier enterprise-grade intelligence using LLMs has traditionally been prohibitively expensive and resource-hungry, and often costs tens to hundreds of millions of dollars. As researchers, we have grappled with the constraints of efficiently training and inferencing LLMs for years. Members of the Snowflake AI Research team pioneered systems such as ZeRO and DeepSpeed, PagedAttention / v
Today we’re excited to announce our Release Candidate (RC) of TypeScript 4.8. Between now and the stable release of TypeScript 4.8, we expect no further changes apart from critical bug fixes. To get started using the RC, you can get it through NuGet, or use npm with the following command: npm install -D typescript@rc You can also get editor support by Downloading for Visual Studio 2022/2019 Follow
The inside story on Mountpoint for Amazon S3, a high-performance open source file client | Amazon Web Services
AWS Storage Blog The inside story on Mountpoint for Amazon S3, a high-performance open source file client UPDATE (8/9/2023): Mountpoint for Amazon S3 is now generally available. For details, please read the What’s New post. Amazon S3 is the best place to build data lakes because of its durability, availability, scalability, and security. Hundreds of thousands of data lakes are built on S3, storing
40 Years on, PAC-MAN Recreated with AI by NVIDIA Researchers
40 Years on, PAC-MAN Recreated with AI by NVIDIA Researchers GameGAN, a generative adversarial network trained on 50,000 PAC-MAN episodes, produces a fully functional version of the dot-munching classic without an underlying game engine. Forty years to the day since PAC-MAN first hit arcades in Japan, and went on to munch a path to global stardom, the retro classic has been reborn, delivered court
Ubuntu Weekly Recipe 第750回LXDとUbuntuサーバーで、RAID障害時の再構築の予行演習と監視について確認する 今回はmdraidを使ったソフトウェアRAID環境における「再構築（リビルド）」を試してみましょう。RAIDにおける再構築とは、簡単に言うと「ストレージ障害の回復作業中にもう一台を壊す、もしくは実は壊れていたことに気づく」仕組みです。きちんと運用されていたら、壊れることになくシステムが回復する可能性は高いのですが、実際のところは作業者のKarma値ないしLuck値次第となります。1%の確率で起きる事象を100%ひいてしまう人が世の中にはいるのです。 そこで今回は少しでも再構築作業がスムーズに進むように、その予行演習を行いましょう。今回もLXDで構築した仮想RAID環境を利用します。LXDを用いてソフトウェアRAID環境を構築する方法は、次の回を参考にし
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
An oral history of Bank Python
Exclusive: A Deeper Look at the PlayStation 5
Welcoming Android 10!
Network Architecture Design for Microservices on GCP
［C++］volatile修飾子についての考察 - 地面を見下ろす少年の足蹴にされる私
Announcing Swift Algorithms
データから得られた知見は実在集団に当てはまるか？外的妥当性について少し深く考えてみた - Unboundedly
ControlNet image sequence 2 image sequence script v2
Font size is useless; let’s fix it
ECSS — Simple Rules for efficient CSS
LCM-LoRA×DiffusersでリアルタイムAIお絵描きを試してみる
U-Netを識別器に！新たなGAN「U-NetGAN」を解説！ - Qiita
CNCF Platforms White Paper
Snowflake Arctic - LLM for Enterprise AI
Announcing TypeScript 4.8 RC - TypeScript
第750回 LXDとUbuntuサーバーで、RAID障害時の再構築の予行演習と監視について確認する | gihyo.jp