Sabitlenmiş Tweet
Layouts as Integer Set Relations - Can cover CuTE and Triton Linear Layouts in common framework more to come!
arxiv.org/abs/2511.10374
English
Vinod Grover
1.3K posts
Vinod Grover
@vinodg
Sr Distinguished Engineer @nvidia. Compilers, CUDA C++, PL, Machine Learning and Systems. tweets and opinions are personal.
Seattle, WA Katılım Mart 2008
1.3K Takip Edilen4.7K Takipçiler
2 Claude Codes went to a bar …
Dimitris Papailiopoulos@DimitrisPapail
launched 2 Claude Codes in separate terminals and told both: "find each other and build something together." They built a prog language in 12 minutes with zero human intervention. Its core feature is collaboration. they named it Duo. A slide from their prez about it
English
Vinod Grover retweetledi
Joe Halpern passed away Friday at the age of 72. He was a leader in mathematically reasoning about knowledge.
linkedin.com/posts/daphne-k…
English
Compiler research intern (Summer 2026) positions at NVIDIA (Seattle) for ML. Functional programming, polyhedral compilation, program synthesis, compiler optimization. Send resume to compiler-jobs@Nvidia.com
English
Vinod Grover retweetledi
mini-SGLang ⚡️ is tvm-ffi native.
It leverages tvm-ffi for AOT & JIT codegen for:
- custom CUDA kernels
- communication primitives
- symbolic tensor matching
- PDL kernel launches
tvm-ffi offers the convenience to freely switch across languages.
LMSYS Org@lmsysorg
How long have you been "planning to understand" how modern LLM inference works? We just gave you a readable version of SGLang you can finish over the weekend. Introducing mini-SGLang ⚡ We distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance — nearly identical to SGLang online. It is built for engineers, researchers, and students who want to see how inference really works and learn better from code than papers. ⭐ Star us on GitHub: github.com/sgl-project/mi… 🧵 (1/3)
English
Vinod Grover retweetledi
Today, @NVIDIA is launching the open Nemotron 3 model family, starting with Nano (30B-3A), which pushes the frontier of accuracy and inference efficiency with a novel hybrid SSM Mixture of Experts architecture. Super and Ultra are coming in the next few months.
English
It was great fun to present and meet people. Thanks to the organizers!
Dan Zhang @ NeurIPS@DZhang50
After leading the creation of CUDA C++, Vinod Grover @vinodg (Sr. Distinguished Engineer, @NVIDIA) now focuses on the compiler techniques that make modern AI training and inference possible. Come listen to his invited talk, titled: "The Essence of CUDA and AI for GPUs"!
English
This one of the projects I contributed to.
The work of many.
NVIDIA AI Developer@NVIDIAAIDev
The largest advancement of the CUDA platform since its creation in 2006 is here 👀 Introducing CUDA Tile, a tile-based programming model that provides the ability to write algorithms at a higher level and abstract away the details of specialized hardware, such as tensor cores. Read the technical blog 👉 developer.nvidia.com/blog/focus-on-…
English
@samiramanabi Sometimes putting more time on a task than it needs doesn’t help. 😀 sometimes there is a better way even if it sounds like it is complicated and time-consuming
English
Both agree and disagree 😂 My experience in academia has taught me that people who are championing work-life balance clearly has the agenda of getting ahead of others.
But at the same time, not sleeping for a few days only make you crash later. It is not exactly how you optimize if you want to sustain long working hours year after year.
Ayush Jaiswal@ayushjaiswal
Work life balance is great, in-fact we recommend it to all our competitors. Don’t be angry with this post. You’re free to disagree & let us cook :)
English
Vinod Grover retweetledi
CuteDSL 4.3.1 is here 🚀 Major host overhead optimization (10-40µs down to a 2µs in hot loops_, streamlined PyTorch interop (pass torch.Tensors directly, no more conversions needed) and export and use in more languages and envs. All powered by apache tvm-ffi ABI
English
Vinod Grover retweetledi
Today we are releasing our first public beta of Nsight Python! The goal is to simplify the life of a Python developer by proving a pythonic way to analyze your kernel code!
Check it out, provide feedback!
Nsight Python — nsight-python docs.nvidia.com/nsight-python/
English
@alpha_convert 90 % of modern stuff was done in the 70s or before. Polished later 😀
English
Vinod Grover retweetledi
Xavier Leroy’s new book “ Control structures in programming languages: from goto to algebraic effects”
discuss.ocaml.org/t/book-draft-c…
English
Vinod Grover retweetledi
📢Excited to introduce Apache TVM FFI, an open ABI and FFI for ML systems, enabling compilers, libraries, DSLs, and frameworks to naturally interop with each other. Ship one library across pytorch, jax, cupy etc and runnable across python, c++, rust tvm.apache.org/2025/10/21/tvm…
English
Vinod Grover retweetledi
Checkout how speculative decoding and #XGrammar can work together to get efficient and accurate structured outputs. github.com/NVIDIA/TensorR…
English
lol, there is quite the explosion of kernel DSLs lately (triton, tilelang, gluon, TLX, cuteDSL, cuTile, …)
And honestly as much as I love TLX and want it to succeed, I think the next big kernel programming language might be… natural, human language
SzymonOzog@SzymonOzog_
Just one more DSL bro. I promise bro just one more DSL and we'll fix hardware adoption. It's just a better DSL bro. Please just one more. One more DSL and we'll port all the kernels. I just need one more DSL
English
@hyhieu226 technically a map operation can be expressed using reduce or a fold. the attached from cambridge.org/core/services/…
English
@hyhieu226 the earliest use of `reduce` (the / operator) originates in Iverson's APL in 1962. functional programming sometimes uses that term but `fold` is more common.
dl.acm.org/doi/pdf/10.114…
English
Hot take: please call my English bad, but I don't understand why map (in map-reduce), chunk, tile, partition, etc. mean the same thing (dividing a large workload to smaller workers); also why reduce (in map-reduce) means to combine? They are not the best word choice. 😒
English