Junru Shao
650 posts
Junru Shao
@junrushao
opinions are my own
California, USA เข้าร่วม Ekim 2012
517 กำลังติดตาม1.8K ผู้ติดตาม
I have gone from the most AI pilled person in the world to a straight up boomer in like a week. The slop keeps slopping and I feel people should have to take a test before they are allowed to use claude
English
🚀 New blog: Elastic EP in SGLang: Making Large-Scale MoE Inference Resilient
Wide EP is essential for serving massive MoE models like DeepSeek-V3, but it also creates a reliability trap: one GPU fault, and the entire instance crashes.
Together with the Mooncake Team & Volcano Engine, we introduce Elastic EP:
⚡️ <10s recovery from rank failures (vs. 2–3 min full restart) — a 90% reduction in downtime
🧠 Zero static overhead: throughput matches standard DeepEP under static conditions
Try it now with SGLang!
English
hey friends! 👋
Only cool people are allowed to reply to this tweet obviously.
English
OpenAI acquired Astral, the team behind uv, ruff, and ty.
Fun fact: Claude is the #6 contributor to uv.
Curious if Anthropic will ban them from using Claude since the team is joining OpenAI.
Congrats to the Astral team who built incredible Python tools!
English
Guys, SF is a magical city.
Today, a beautiful Wednesday, I had a 4:30pm meeting with a friend at a café… it was closed.
We walked 10 blocks, every café was closed. Finally found Blue Bottle, it closed at 5:30.
Can some YC company please build what SF people actually want?
English
Yuchen is the smartest and the most capable person I’ve ever worked with in my career. I’m deeply saddened to see how this unfolded, and wish @Yuchenj_UW the very best 🙏
Yuchen Jin@Yuchenj_UW
I have decided to step down as CTO at Hyperbolic. Leaving a company you co-founded and poured your heart into is not easy. So many moments still feel vivid: launching our AI inference product for open-source models and seeing tens of thousands of developers sign up in a week; the week we were hit by a massive DDoS attack and the entire engineering team fought around the clock until we won; the day we launched the GPU platform and watched ARR take off. There were also hard moments. That’s the nature of building a startup. I’m grateful for all of it. What I’m most grateful for is the team. Thank you for your trust. Most startups never build something people want. I believe we did. You should be proud of yourselves. I will look forward to seeing your success. What’s next for me? I’m still figuring it out. I believe this is the most extraordinary moment in human history. We’re standing at the edge of the Singularity. AI will reshape everything, and I still feel the same excitement I felt when I first fell in love with AI. Time to start over. Time to climb another mountain. Thank you to everyone who has been part of the journey, — Yuchen
English
might be an ego problem but anyone really hate when someone use the "request changes" on their PR?
English
Introducing the Google Workspace CLI: github.com/googleworkspac… - built for humans and agents.
Google Drive, Gmail, Calendar, and every Workspace API. 40+ agent skills included.
English
@henrylhtsang sounds like a good idea actually😅 or vibe code a paste bin server lol
English
@junrushao I was thinking something more hacky, like have agent 1 to post to github gist, and agent 2 check every 5 minutes
English
what is the best way to make agents from different server to talk with each other
English
Junru Shao รีทวีตแล้ว
Layouts as Integer Set Relations - Can cover CuTE and Triton Linear Layouts in common framework more to come!
arxiv.org/abs/2511.10374
English
黄老板所说的 “买得越多，省得越多” 真的靠谱么？
3月7日，GPU Mode 第100场激情拆解。黄老板亲传子弟 Dylan Patel，基于 @SemiAnalysis 全新 InferenceX 基准，直播拆解 GB300 NVL72（假设你能买到）和 H100 的真实推理性能天花板，扒一扒核弹厂的底裤。
顺便，InferenceX 实锤：老黄的刀法，确实比苏妈的芯片更锋利 🔪
算力军火商的私房课，来了
中文
Junru Shao รีทวีตแล้ว
This is why nmoe and all of my training codebases are TP free. Wide EP + NVlink is the answer
Federico Cassano@ellev3n11
tensor parallelism is a scam
English
@henrylhtsang Claude’s analysis is surprisingly accurate lol github.com/Dao-AILab/flas…
English
@shl gh is where my coding agents address review comments realtime
English
@roeschinc Claude code is now doing slack design debates for me. I feel myself so useless lol
English
I know the "inflection point" tweets have been going so strong about agentic coding the last few months, but over the last few weeks I feel like they have finally hit a level of capability where they can productively work with little intervention in code bases like compilers.
It has been an insanely fun/interesting experience to watch me and another engineer implement the majority of a large feature while writing almost no code directly, and mostly working on design docs, and PLAN.md files.
This has all been on Opus-4.5/4.6 in Claude code. I am starting to get excited about being able to evolve how we do engineering at scale using these tools!
English
Junru Shao รีทวีตแล้ว
The age of agent makes more demands for people think think like an architect rather and move away from ground works, but without experience gained through ground works, can one become a solid architect, it is a fun question and challenge for students in the new age.
English
Junru Shao รีทวีตแล้ว
V4 Lite now live in the app.
1M context length. Text-only.
Muon + mHC confirmed.
Larger version is still on the way.
@zephyr_z9 @teortaxesTex
English