Varad Rane

inference at @simplismarthq

[Varad Rane @varadrane17]

@maharshii Was fun winning the buildathon as well. Nice talk on the quant and GEMM kernels btw

[Varad Rane @varadrane17]

@0xAmbesh @sama Blocked you for ....

[Varad Rane @varadrane17]

Looking for the technical report

[Black Forest Labs @bfl_ai]

Introducing FLUX 3.

One multi-modal model for Image, Video, Audio and Action-Prediction. Creations are truer to life in every kind of style.

FLUX 3 Video is now available in early access (link below).

Jointly trained in one unified architecture, our model can be extended to predict actions for robotics. See our work with mimic and Audi in the thread.

[Varad Rane @varadrane17]

You need patience to learn kernels.

[Varad Rane @varadrane17]

@jino_rohit More i think since most LLMs are compute bound and we see less improvements in real production traffic when we have variable concurrency and qps?

[ambesh @0xAmbesh]

Was going through the Iceberg Summit 2026 recorded talks and hit a great one:
how Apache Iceberg moved from v2 merge-on-read deletes → v3 deletion vectors.
https://t.co/g5s19Degex

>In v2, deleting a row meant writing side "delete files" (positions or matching values) that the engine had to merge back in on every read and they pile up into lots of small files that slow reads down.

>v3 swaps that for a deletion vector: one compact Roaring bitmap per data file, bit P set = row P deleted.

That sent me down a rabbit hole into the paper behind it — "Better bitmap performance with Roaring bitmaps" (S. Chambi, D. Lemire, O. Kaser, R. Godin). Here's my deep dive 👇
https://t.co/o8z6ZEhExm

[Tianqi Chen @tqchenml]

We taught a brand-new mini-series this year at @SCSatCMU on Modern GPU Programming for ML Systems, as part of the ML Systems course, touching on fun questions like what data layout swizzling is, how to use 3D TMA, and state-of-the-art Blackwell programming. We released a curated online book based on the materials: https://t.co/5ZJg2lySNO check it out

[Varad Rane @varadrane17]

@maharshii Hell yes!

[Varad Rane @varadrane17]

@maharshii Which model? Opus 4.8 or Fable?

[Varad Rane @varadrane17]

@DanielSmidstrup Cloud Cost

[Varad Rane @varadrane17]

@0xAmbesh Claude Fable 5 - 2026-2026 RIP 🪦

[Varad Rane @varadrane17]

@prathamgrv @TensorTonic Looks amazing , does it have support for CuteDsl or triton as well?

[maharshi @maharshii]

You can easily get intellisense and auto-complete for nvidia-cutlass-dsl with vscode on Macbook for kernel development purposes:

> download "nvidia cutlass dsl libs base" wheel
> not the "nvidia cutlass dsl" wheel, it does not contain any python source
> extract it to a directory
> get full path for "nvidia_cutlass_dsl/python_packages" from the extracted directory
> add it to pylance in vscode under "python.analysis.extraPaths"

Note: for actually testing the code, you will need a nvidia GPU to run it on. I figured this out just now, so there might be some things that I may have overlooked.

[Varad Rane @varadrane17]

@sahill_og + 289K Tokens used

[Varad Rane @varadrane17]

@tushargoel_ai Lets work at Simplismart Ai ?

[Varad Rane @varadrane17]

I love how it see is so easy to publish tech blogs on https://t.co/458KJgYggH
FYI First Blog on BearBlog : https://t.co/Ha1qJhyoJ5

[Varad Rane @varadrane17]

@sama Maybe better quality like Anthropic models ?

[Varad Rane @varadrane17]

Is it me or does anyone else find no good tutorial for CuteDSL-python ?
#nvidia

[Varad Rane @varadrane17]

@NoahKingJr Jarvis ?

[Varad Rane @varadrane17]

@insanekrishnaa hi interested