2024.06.10 - Beginner's Guide to Large Model Fine-tuning and Deployment
Ping Xia
Title: 2024.06.10 – A Beginner’s Guide to Fine‑Tuning and Deploying Large Models
The Lost Web & Local‑first & Shao Yibo
This Week’s Highlights
A Beginner’s Guide to Fine‑Tuning and Deploying Large Models https://zhuanlan.zhihu.com/p/666976486
Compared with RAG, fine‑tuning a large model requires some knowledge of neural networks and has a higher entry barrier. This article aims to introduce those basics so ordinary developers can fine‑tune their own models. Also included:
- Extracting Concepts from GPT‑4
- NotebookLM goes global with Slides support and better ways to fact‑check
- A Picture is Worth 170 Tokens: How Does GPT‑4o Encode Images?
- What We’ve Learned From A Year of Building with LLMs
Alibaba Has No Product
https://mp.weixin.qq.com/s/2ctGJEtnwj83iUo0XfNoFg
Alibaba is impressive on many fronts: it runs online commerce with astonishing efficiency, maintains long‑term relationships with merchants, reshapes our consumption habits, and boasts strong execution and sales capabilities. None of that is denied.
However, I think Alibaba’s biggest problem is that it remains a “sales‑first” company that doesn’t care much about product quality or user respect. It shows the best traits of traditional Zhejiang merchants—fluent speech, down‑to‑earth, meticulous with money, respectful of customers—while also displaying the less admirable side: little product focus, user disregard, arrogance, and overconfidence.
In Search of the Lost Web https://maraoz.com/2024/06/05/in-search-of-the-lost-web/
Some of you remember the 2000s with nostalgia—the “good old days” of the Internet. Back then, most Google results led to a real human’s blog. YouTube had no recommendation algorithm. Social media didn’t exist. People behaved oddly online instead of posing. The web felt homey, not an endless ego contest where everyone tries to prove how sophisticated they are. Interest‑based forums were genuine communities; many of us made lifelong friends there. Today’s web feels dry, over‑polished, dramatic, and soulless—all at once. What can we do? Related: Beware the cloud of hype.
Home‑Cooked Software and Barefoot Developers https://maggieappleton.com/home-cooked-software
The emerging golden age of home‑cooked software, barefoot developers, and why the local‑first community should help build it.
In‑Depth Reading
How GitHub reduced testing time for iOS apps with new runner features https://github.blog/2024-06-03-how-github-reduced-testing-time-for-ios-apps-with-new-runner-features/
Learn how GitHub used macOS and Apple Silicon runners for GitHub Actions to build, test, and deploy our iOS app faster.
Measuring What Matters: The Science of UX Design https://www.browserlondon.com/blog/2024/06/03/measuring-what-matters-the-science-of-ux-design/
You get what you measure. Measure the wrong thing and you get the wrong behaviours.
Implementing end‑to‑end encryption for Dropbox teams https://dropbox.tech/security/end-to-end-encryption-for-dropbox-teams
Here we’ll discuss our implementation of end‑to‑end encryption for teams, the threat model of our design and encryption algorithms, and our commitment to minimizing the risk of data loss with a team‑centric key management approach.
How Stripe’s document databases supported 99.999% uptime with zero‑downtime data migrations https://stripe.com/blog/how-stripes-document-databases-supported-99.999-uptime-with-zero-downtime-data-migrations
In this blog post we’ll share an overview of Stripe’s database infrastructure, and discuss the design and application of the Data Movement Platform.
Redesigning Pinterest’s Ad Serving Systems with Zero Downtime https://medium.com/pinterest-engineering/redesigning-pinterests-ad-serving-systems-with-zero-downtime-3253d2432a0c
In this blog post, we will describe the motivations and challenges of this rewrite, along with our wins and learnings from this two‑year journey.
What do I think about Lua after shipping a project with 60,000 lines of code? https://blog.luden.io/what-do-i-think-about-lua-after-shipping-a-project-with-60-000-lines-of-code-bf72a1328733?gi=8385ed7c6379
Hi there! This is Oleg from Luden.io. We decided to have a deep and meaningful conversation about the Lua programming language with Ivan Trusov, lead programmer of the video game Craftomation 101. It contains ~60,000 lines of Lua code and is made with the Defold game engine.
Interactive story generation https://storybook.js.org/blog/interactive-story-generation/
Make your first story in seconds without leaving the browser!
Fresh Finds
WebAssembly JSPI has a new API React‑PDF: Easily display PDFs in your React app. Motion Canvas – a TypeScript library for creating animated videos using the Canvas API. ASCII Silhouettify: an app that converts images into ASCII silhouettes Python notebooks for fundamentals of music processing QStudio 3.0 – the leading SQL editor for the modern data analyst
Products & Others
You and I are both Shao Yibo
https://mp.weixin.qq.com/s/0ri8v0VXjRUEOdsHnuqBmA https://mp.weixin.qq.com/s/unM3IxnBH5PLJjqU2TKVjQ
For many modern people, happiness and peace have become unattainable mirages; making money at all costs seems the only lifelong pursuit. Yet perhaps life shouldn’t be like that. The world is indeed full of suffering, but the more we suffer, the more we should instinctively strive to overcome pain and seek happiness—that should become our deepest instinct and ultimate goal. It’s like Earth’s gravity: resisting gravity and growing upward is a biological instinct. As the “spirit of all things,” we even stand upright on two legs, build airplanes to soar, and craft spacecraft to walk in space—abilities unique to humans that defy gravity. We should meet pain the same way: grow opposite to its pull, shed its weight, and extend freely toward the spiritual sky. In other words, look inward, know yourself, become aware, develop spirituality, and walk the path of awakening. More and more people will take this road; Shao Yibo simply stepped onto it a little earlier.
When you feel empty and exhausted, how can you find nourishment and comfort in ordinary life? https://mp.weixin.qq.com/s/fRQ7gb9pJVYo_6I-IPoLmQ
You constantly feel the house isn’t big enough, the wardrobe isn’t full enough, the partner isn’t attentive enough, the child isn’t as outstanding as others’. Anxiety and a sense of lack fill you, as if a hole needs filling with countless things, yet you don’t know what or how much. In this endless chase for external things, you accumulate more possessions, titles, resources, while the inner self becomes increasingly tired and hollow. How can a weary, anxious heart finally settle down?
Summer in Jiangnan https://mp.weixin.qq.com/s/Tlsachylc40ltNNcTmYiOA
When summer arrives, Jiangnan sheds its modesty and gentleness, revealing a different scene. Some say summer nostalgia in Jiangnan is a sea of lush trees, chorus of frogs, vibrant green heat, endless lotus leaves, shade‑draped trees, mountain breezes, and cool well water. It is this Jiangnan that adds poetry and surprise to the sweltering season.
“National Health Conference” – tears and laughter! https://mp.weixin.qq.com/s/0DhheDym7ziNt2v755vaIw
“Prevention is the most economical and effective health strategy.” Health links to the happiness of millions of families and the future of the nation. The Central Radio and Television Station’s Social Education Program Center has poured effort into creating the National Health Conference. An authoritative medical expert team forms a “Health Science Popularization Super‑Group” to address the public’s most pressing concerns, delivering practical knowledge that is “easy to see, easy to learn, easy to use”; it uses highly imaginative ways to roam the human body, making health concepts vivid, visual, and tangible.
A single Chinese herb can drill into bone cracks and, like a high‑pressure water gun, flush out phlegm and dampness completely! https://mp.weixin.qq.com/s/W2in80AvU2GP7jOIRAFaPQ
Speaking of eye bags, earlier this year I met a patient who had them for six years, tried salons and hospitals, tried many methods, but nothing worked; after a temporary fix the bags returned even worse. Referred by a friend, he came to me. After examining him, I wrote the award‑winning prescription: Xiao Shen Jin Cao and Bai Jie Zi.
-- THE END --
Originally written by Ping Xia (平侠) and published in Chinese on Web技术周刊 (Web Tech Weekly). Translated and adapted for DriftSeas with permission.
Sources & References
- [1]https://zhuanlan.zhihu.com/p/666976486
- [2]Extracting Concepts from GPT‑4
- [3]NotebookLM goes global with Slides support and better ways to fact‑check
- [4]A Picture is Worth 170 Tokens: How Does GPT‑4o Encode Images?
- [5]What We’ve Learned From A Year of Building with LLMs
- [6]https://mp.weixin.qq.com/s/2ctGJEtnwj83iUo0XfNoFg
- [7]https://maraoz.com/2024/06/05/in-search-of-the-lost-web/
- [8]Beware the cloud of hype
- [9]https://maggieappleton.com/home-cooked-software
- [10]https://github.blog/2024-06-03-how-github-reduced-testing-time-for-ios-apps-with-new-runner-features/
- [11]https://www.browserlondon.com/blog/2024/06/03/measuring-what-matters-the-science-of-ux-design/
- [12]https://dropbox.tech/security/end-to-end-encryption-for-dropbox-teams