🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

discovered a pretty cool beta feature of Modal in this cabinet at HQ 🤫

it's funny how people around you notice emotional gravity/weight before you do yourself. like if someone has known you for 3 years: maybe when they say you're down, you might actually be not having the best day even if you think you're fine

A quick story: when I first started doing engineering at Modal we had a little thing that let you write code locally, and it runs on the cloud. This was kind of exciting to me, and so I joined this random startup with 0 revenue, and started writing code. Now it's over 3 years

[jax-js] So I definitely underestimated the complexity a bit. But a few compiler/graph algorithms later, I have a working version of jax.jit() This means that we have a fully end-to-end, optimized jax.numpy.matmul() now, with kernel fusion + autograd + vmap support :D

It's lowkey a bit strange now when I meet people for work reasons. Years ago, I used to avoid it like the plague since I had a complex abt people around me only seeing me transactionally; but time passed, I left harvard bubble, and my relationship to work gradually changed too

OK this is organized now! We're having NY systems reading group in SF, this Tuesday. Come swing by! We'll have 20-35 people and take some quality time to learn / discuss, teach each other a bit. Topic is reading the Firecracker VMM source code lu.ma/6t4edg1d

Here's the distribution of responses I got to survey question I asked to everyone who signed up for systems reading group in SF! (on Firecracker) Lots of ML systems, traditional distributed systems, and data systems (incl DBs)— less networks/graphics/compilers stuff

best part of making a new product is seeing your coworkers be the first avid users :D thank u Charles 🎉 Frye

first SF culture shock is people sending me gcal invites to things we're scheduling in-person? which is honestly great but also I'm just kind of surprised :')

NYSRG in June 2025 will be reading a bit about modern vision systems. For some context why I think it might be a good time. This will be kind of catching up to the state-of-the-art, what's changed over the past 5 years in deep learning, how do the models work (mathematically and

Thanks for coming to NYSRG in sf!! this was the craziest and biggest event I’ve organized so far and it was awesome reading Firecracker source code with y’all. So many incredible people in SF and much smarter than me. Thank you Shreya Shekhar (Greylock) for the space / hosting

Spotted in South Park! SF is telling me to keep hacking on JAX :)

I think I've actually fully exhausted my work+social battery, and it's only been 3 days in SF! Might need to scale it back a bit so I still have something left / can also keep going to work every day :D I think I talked for 19 hours over the past 3 days omg

Rolldown-Vite is ready for you to take it for a spin! Still lots of work to do, but we are seeing very promising stats in real-world applications. Help us battle test it and get it stable sooner!

I have made my pilgrimage to the holy ground of cloud systems research 🫡

it is a little sad when you catch up with friends after a couple years and ask them about their life, but they don't reply and ask how you're doing in turn. but this is okay, maybe someone just had a crazy dense few months, and you're a bit of relief for them – a breath of air

nysrg reads about modern image systems! today I spent some time catching up on score matching since 2020, and nerded out about flows with the astrophysicist-turned-quant sitting next to me. overheard others reading about ggml, decompiling the Zoom app, etc

We've run thousands of LLM inference serving benchmarks at Modal. We're releasing the results so you don't have to. We're releasing the code so that you can. Introducing: The LLM Engineer's Almanac. Just in time for the AI Engineer World's Fair.