Thai Nguyen Trong

#FinoAllaFine ⚪️⚫️

[Amit Shekhar @amitiitbhu]

Learn LLM Inference Engineering step by step - from KV cache, PagedAttention, and continuous batching to vLLM, SGLang, and GPUs.

https://t.co/89bnkj5M7S

[Amto @XAMTO_AI]

去年想搞清楚LLM到底怎么工作的，翻了不少教程，要么只讲概念不写代码，要么直接上transformers库把底层全封装了，看完还是懵。

这个项目（Train LLM From Scratch）用纯PyTorch从零实现Transformer，没有trl、peft、transformers这些依赖，
每一步都拆开给你看：数据怎么变成token、模型怎么搭（MLP→注意力→多头→Transformer块→完整模型）、预训练怎么跑、SFT怎么教模型回答问题、Reward Model怎么打分、DPO/PPO/GRPO怎么对齐，整个链路清清楚楚。

代码里每个模块都有解释和预期输出，单GPU就能跑百万到十亿参数的模型，硬件参考也标得挺细。

🔗 地址：https://t.co/VXF7XOsoBb

[mousepotato @iluciddreaming]

https://t.co/SyjLrlHBWc

[AVB @neural_avb]

Repo here: https://t.co/Qvb9BLNXmT

[harrrshall @HarshalsinghCN]

I built an open-source Hinglish TTS that beats every model on the market. I had zero research background.

last week I was using Ira AI companion. Got curious about the company. Started digging into their stack.

I set a stupid goal to build the world's best Hinglish TTS from scratch.

Documented every dead end, every architectural decision, every mistake and how i approached this.

Full build log (open source): https://t.co/UA6Lh3LDQ7

@aiweekendsxyz @blrwritersroom

[灰狐 @huihoo]

《Introduction to Modern Statistics》Second Edition
Free Math Book, 510 pages, CC License. This book is available as a PDF and in paperback. also read the book online
https://t.co/Zo90LRqMAO

A college-level open-source textbook with a modern approach highlighting multivariable relationships and simulation-based inference

Hope readers will take away three ideas from this book in addition to forming a foundation of statistical thinking and methods.

1、Statistics is an applied field with a wide range of practical applications.
2、You don’t have to be a math guru to learn from interesting, real data.
3、Data are messy, and statistical tools are imperfect. However, when you understand the strengths and weaknesses of these tools, you can use them to learn interesting things about the world.

Part 1: Introduction to data.
Part 2: Exploratory data analysis.
Part 3: Regression modeling.
Part 4: Foundations for inference.
Part 5: Statistical inference.
Part 6: Inferential modeling.

This is a book from the Open Textbook Initiative Approved Textbooks.
https://t.co/MpnR4ID16P

[Joruno @wsl8297]

现在大家都在做 Agent，但一上复杂业务：架构怎么搭、记忆怎么管、多智能体怎么配合，往往越做越迷糊。

我最近看到一本开源书 Agentic Design Patterns，把智能体设计模式从入门讲到企业级，脉络清晰、拆解到位。

全书 21 章 + 7 个附录，按难度分四大部分；每章配套 Jupyter Notebook，边读边跑，理论和代码紧贴在一起。

GitHub：https://t.co/MqH1Cmo5oL

前半段打底：提示链、路由、并行、反思、工具调用、多智能体协作等核心模式，一次讲透“怎么设计”。

后半段落地：记忆管理、异常恢复、人机协作、安全护栏、性能评估等生产必修课，直接对准“怎么上线”。

附录还补齐框架对比和高级提示技巧，适合查漏补缺、随用随翻。

想把 Agent 从概念学到可落地的系统方法，这本开源书值得收藏，慢慢啃。

[Felix Heide @_FelixHeide_]

Are we done with object detection? What about tiny objects beyond 200 meters? 🔎

Telescope 🔭 addresses long-range perception by explicitly tackling extreme scale imbalance ⚖️ in images. It hinges on a learnable hyperbolic foveation transform from a low-resolution image, magnifying distant regions 🔍 while compressing nearby ones - effectively normalizing object scales with minimal computational overhead. Objects are detected in the transformed (Riemannian) space using a novel bounding box parameterization and are then mapped back to the original image.

Project: https://t.co/mBuQGd7KnB

[Tom Dörr @tom_doerr]

Private, multi-model, local alternative to Notebook LM

https://t.co/cwQJXHuyiJ

[Vaishnavi @_vmlops]

This might be the wildest AI engineering breakdown on the internet right now 🤯

After the Anthropic leak…

Someone turned the ENTIRE Claude Code system into a readable playbook.
👉 https://t.co/d6OptOkWbC⁠

We’re talking:
* 500K+ lines of real production AI agent logic
* Broken down into 18 chapters you can actually learn from
* Multi-agent systems, tool pipelines, memory, orchestration… all exposed

This isn’t theory
This is how a top-tier AI coding agent actually works under the hood

Key ideas you’ll steal instantly:
→ Agent loops with async execution
→ Multi-agent “teams” coordinating tasks
→ File-based memory (no DB 🤯)
→ Context compression tricks
→ Tool execution pipelines at scale

Basically…
Instead of guessing how to build AI agents
you now have a blueprint from a real system used by thousands of devs
https://t.co/8R2lBSfn3C

Crazy part?
The whole thing was analyzed + rewritten in HOURS using AI agents
https://t.co/8R2lBSfn3C

If you're building:
• AI agents
* Dev tools
* LLM products
* or learning MLOps

This is not optional
This is a cheat code

[Thai Nguyen Trong @gin97_]

@JuveGiovanili Solid on the ball but a bit slow in transition. Still he’s got that confidence to make something happen

[Thai Nguyen Trong @gin97_]

@DaltonCassar @NicoSchira Nobody does it better than him. Who else should we consider?

[Thai Nguyen Trong @gin97_]

@thomas_novello9 Pellegrini

[Thai Nguyen Trong @gin97_]

@Jfamily3502 He cant if the corrupt system still exists

[Thai Nguyen Trong @gin97_]

@gualanodavide Great stuff! Juve faced Stuttgart recently so Stiller is definitely on their radar. Heard rumors they’re scouting English-based midfielders like Tyler Morton, Hayden Hackney... What’s your take on them?

[AlphaCephei @alphacep]

Finally a good VAD

https://t.co/CBgtArAEi8

so many developers use bad but popular VAD these days

[Vaishnavi @_vmlops]

Someone literally built a free AI university - all in one repo, covering real-world AI systems step by step

Link - https://t.co/itaqmWZTBQ

Here’s what you’ll learn:

Week 1 - Setup everything
Docker, FastAPI, databases
Beginner-friendly foundation

Week 2 - Feed it real data
Automatically fetch research papers
Fully automated data pipeline

Week 3 - Teach it to search
BM25 keyword search implementation
Your own Google-like search system

Week 4 - Make it smarter
Hybrid search enabled
Understands meaning, not just keywords

Week 5 - It talks back
Complete RAG system
Ask questions, get accurate answers

Week 6 - Production ready
Caching and monitoring added
Runs like a real product

Week 7 - Give it a brain
Agentic AI with LangGraph
Even works with Telegram

[Yatharth Sharma @Yatharth3501]

Just open sourced LavaSR v2: an incredibly fast model that can enhance 5000 seconds of audio in 1 second.

V2 is much higher quality than a 6gb model now and v1.

It's great to
- Enhance TTS models
- Fix Audio Datasets
- Restore low quality audio.

Repo: https://t.co/7lEgBCBNu7

[Pau Labarta Bajo @paulabartabajo_]

Wanna learn how to fine tune a small model to excel at function calling?

Here's an example with full source code ↓
https://t.co/JhXMathA8T