Very happy to see our algorithm "Selective Knowledge Transfer" (SeleKT) being adopted in the development of state-of-the-art models.
Huge congratulations to the @continuedev team on releasing a SOTA open-source Next Edit suggestion model, and for sharing their learnings in such…
Very happy to see our algorithm "Selective Knowledge Transfer" (SeleKT) being adopted in the development of state-of-the-art models.
Huge congratulations to the @continuedev team on releasing a SOTA open-source Next Edit suggestion model, and for sharing their learnings in such…
I have so many mix feelings for this paper but let's talk.
A clever perspective of seeing SFT objective as an RL objective and making conclusions.
Suppose:
1️⃣Sample potential y from the policy pi(y|x)
2️⃣A binary reward i.e 1 iff (y == y*) else 0
3️⃣To balance the policy sampling,…
This is what I don't like, SM100 (Blackwell arch for datacenter GPUs like GB100/GB200) supports tcgen05 but the SM120 (Blackwell arch for consumer GPUs like RTX 50 series) don't
ref: forums.developer.nvidia.com/t/how-to-load-…
although its for FP8 (but compiler error is "not supported")
Also if…
This is what I don't like, SM100 (Blackwell arch for datacenter GPUs like GB100/GB200) supports tcgen05 but the SM120 (Blackwell arch for consumer GPUs like RTX 50 series) don't
ref: forums.developer.nvidia.com/t/how-to-load-…
although its for FP8 (but compiler error is "not supported")
Also if… https://t.co/d7MQfxvBi3
Here's how to properly benchmark CUDA kernel (inspired from the working of NVBench)
0.0 Use CUDA_MODULE_LOADING=EAGER (default from cuda-12.3)
1. Do the warm up
2. Flush the L2 cache
3. Block the stream before enqueuing the ops and then unblock
4. Only record timings from…
639 Followers 3K FollowingProduct of progressive public policy; raised by public libraries and public education that produced a passion for politics. and apparently alliteration
321 Followers 3K FollowingResearcher in math+formal methods+ml. Working on using formal verification to train models for mathematics and reasoning @harmonicmath
92 Followers 520 FollowingUG Research Assistant at MURGe-Lab w/ @mohitban47, undergrad at @unccs. Interested in LLM Compression, interpretability, and embedded applications
66 Followers 715 FollowingIPTV es una forma sencilla de transmitir un canal de TV usando no cable ni satélite sino Internet, por lo que para esto necesitas listas de IPTV
176 Followers 4K Following“But to you who are listening I say: Love your enemies, do good to those who hate you, bless those who curse you, pray for those who mistreat you🙏🏾😇❤️
1K Followers 1K Following25 | SWE | Technical Writer | Math Wizard | Building Medtech B2B SaaS Startup | Exploring AI & ML | Rust Community Discord: https://t.co/LUBbclBTHz
15K Followers 1K FollowingSenior Research Scientist - @google, Adjunct Faculty - @iitmadras, @iitbombay, Ex: @NICT_Publicity
Use of my tweets without permission ➡️ legal action
321 Followers 3K FollowingResearcher in math+formal methods+ml. Working on using formal verification to train models for mathematics and reasoning @harmonicmath
3K Followers 342 FollowingI’m a software engineer building high-performance kernels and compilers at Anthropic! Previously at Facebook/Meta (PyTorch, HHVM, ReDex)
54K Followers 0 FollowingWe are building a world class AI R&D company in Tokyo. We want to develop AI solutions for Japan’s needs, and democratize AI in Japan. https://t.co/1q07mb3TzE
2K Followers 470 FollowingResearcher, Royal Society Industry Fellow, Senior Lecturer. Working on interpreters, compilation, concurrency, and debugging tools.
721 Followers 157 FollowingNeuroengineer building speech BCIs | BWF CASI & A.P. Giannini Postdoctoral Fellow @UCDavis department of neurological surgery | Previously @PittBioE
2.4M Followers 47 FollowingThe official handle for NVIDIA. Blog: https://t.co/JAn5eKOTBT Support: https://t.co/6ln5FVnA2o All our social media: https://t.co/Uc56dL57Dh
259K Followers 260 FollowingYour leading source for Marvel News.
#MarvelZombies is coming to Disney+ on September 24.
#SpiderManBrandNewDay in theaters July 31, 2026.
17K Followers 20 FollowingA high-throughput and memory-efficient inference and serving engine for LLMs. Join https://t.co/lxJ0SfX5pJ to discuss together with the community!
45K Followers 64 FollowingStudent of mind and nature, libertarian, chess player, cancer survivor. @ Keen, UAlberta, Amii, https://t.co/u8za2Kod54, The Royal Society, Turing Award
16K Followers 529 FollowingTech Journalism with zero ads, & zero Big Tech influence. We cover the Big Tech stories that other publications are afraid to touch.