Remiscus

@elliotarledge @himanshustwts waiting!!!

@himanshustwts hold on a min the article is almost done

K3 is SOTA in GPU Kernel Optimization

@redtachyon 4.1 was good, plus its a 31b open model, what more could you want?

>beats GPT-4.1 Oh alright then. That was... a model. That I remember. Vaguely. It came out sometime last year, I think? It was... decently good for its time, probably?

@modal fucking hell, endgame engineering team you guys got, its honestly amazing to run into a problem like this

@elliotarledge jesus I gotta be you

timelapse #164 (15.5 hrs) - Burning midnight oil with the Kimi team benchmarking on kernels - Getting a Persistent Fused Mega Kernel for a Puffer Lib Environment (the full Craftax game) - Dipped for the night to have models run my overnight training runs on autopilot, including some kernel-based GRPO on qwen3.6 - private contract work - downloaded 1.1 TB of minecraft videos to train a foundation model on - optimized my minecraft foundation model training pipeline and kernels to train on 500k hrs of 128px at 10 fps in ~28 hrs on single gpu (5 hrs of learning per second on rtx pro 6000 blackwell) - doing it for the love of the game

@emilyhanyf @modal Ahhh every company is hiring and baiting me for not being based in US

yes, working at @modal is as fun as it seems. we are casting 35 more roles across new york, san francisco, and stockholm. one could be yours ! modal.jobs

@airkatakana I mean there was only so much they could use to convince us Bag of words and lemmetization is where langauge intelligence at

no one hates language models more than natural language processing researchers who spent their entire lives with the goal of making nlp systems with their deep linguistics knowledge to prove how smart they were only to get bitter lessoned and thrown to the side like trash

@jxnlco I have worked in Voice AI for the better part of my roles and built sybl as a some dev tooling because I was tired of wisprflow

dictation is the first step of a universal voice interface openAI is looking for an engineer to help build the future of dictation in ChatGPT and Codex. If you have experience building these kinds of voice interfaces in the past I'd love to talk to you. Just reply with a link to what you've built.

@BarathAnandan7 Y'all tweet anything for the algo

I’ll give you one piece of advise Grab yourself a 4TB SSD and download it as soon as the weights are out.. IYKYK.. you’re welcome.

@borrasque_ Yeah it all works out if you are based of SF

i only sent 106 cold emails as a college freshman before landing my role as a swe intern at a yc-backed startup what most people don't realize is that sheer volume is not enough i've had multiple friends who've sent 350+ emails yet still got nothing my piece of advice: > be straight forward about what you want > make yourself stand out but be relatable > just because something is "cracked" doesn't mean it's perceived that way think of it as 106 well aimed shots vs 350 blind

@hthieblot I made the same thing, with a being your own key setup, it's just dictation with computer use

This man is 100% the next Steve Jobs

@willccbb Didn't know you guys hired outside SF

this is who you're up against btw

@samirande_ worked for 4 companies during college, applied to like 200 startups in us got 30 rejects rest just ghosted, 4 of them at least told me its not the qualifications but rather that I am not based in US

"I'm gonna complete MERN stack in 6 months and get a $100k USA based remote job"

@gary_miklos @levelsio Commit every loc

I was pretty satisfied with Fable doing 100+ commits a day for me since I was doing maybe 10–20ish before. Then checked @levelsio’s stats and 1,000+ commits in 2024?? and 500+ in 2022, pre-AI? How???

@fredrikalindh If you are looking for a job, how do you pay for tokens?

if i were looking for a job today i’d prioritize working with people who are agent maxxing and have unlimited tokens. you want to position yourself to become extremely good at working with agents and the only way to do that is to do it a lot

@Tanishqstwt Does your side project have 80k users? Why kafka?

i love building my side projects with go, grpc, kafka, k8s and helm.

@waterloo_intern @nvidia Damn... Is this what it takes to be an intern nowadays? Anyways waiting on the paper, I touched this problem a little bit didn't have any meaningful compute to test things on, so it didn't go anywhere

it took 1 intern 3 months of continuous work, but eventually, a quantization method that beat every other algo in the market, including @nvidia's official modelopt to explain why this matters, i ask for exactly 69 seconds of your attention (275 words @ avg reading speed of 238 wpm): frontier models (like glm52) are huge (~0.8T params). as released, each parameter takes 2 bytes (bf16), so overall size is about 1.6 tb a b200 has 180gb of memory. a node of 8 gives you 1.44 tb, barely fits weights, much less activations / kv cache must quantize the model (reduce the size of each individual parameters) to serve. fp8 quantization means each parameter takes 1 byte (fits in 0.8 tb), fp4 takes 1/2 a byte (fits in 0.4 tb) cutting the model to a quarter its original size is necessary for it to run a) cheap b) fast, and every lab serving models does this. but, quantization lobotomizes the model if not done correctly (this is why you see people complain about @AnthropicAI nerfing claude or @OpenAI nerfing codex) there are currently several algorithms (like Nvidia's official model-opt) that attempt to figure how to quantize a model with the least amount of damage. they find the redundant layers that can be slashed, and sensitive/important layers that need to stay in full-precision. these algo's have two drawbacks: 1) they take a long time to run 2) they quite often result in a sub-optimal configuration for the past 3 months, a research (and, as always, waterloo) intern on our model perf team (@the_joshua_hill) came up with a new quant algorithm. it consistently finds the optimal configuration: a) in less time than SOTA b) with more aggressive quant than SOTA c) scoring higher on benchmarks than SOTA achieving just one of the above is a feat on its own. all three...excited for the paper to come out this week

Guess no one on linkedin is calling themselves prompt engineers anymore huh?

@leo1yu Blessed with resources and a great team, envy you but also happy

one of our senior engineers took PTO to attend his own high school graduation 💀

@pydantic you guys gotta do better PR! day one here, so much better than langchain langgraph or whatever else

Pydantic AI version 2.8.0 is out! 🎉 github.com/pydantic/pydan…

@0xSero Man you circle is fucking amazing

GLM-5.2 Nvidia NVFP4 No pruning, no quantising, exact same model card. 110 tok/s single stream. My friend is a genius, holy fudge. Here's the repo, he got full DS4-Flash on a 5090 + DDR5 github.com/kacper-daftcod… at 38.5 tok/s This is a breakthrough.