Vikram @msharmavikram
@NVIDIA Sr. Research Scientist Large Scale AI/ML Systems | Ph.D. (UIUC - Prof Wen-mei Hwu) All opinions and tweets are personal. msharmavikram.github.io Joined November 2008-
Tweets334
-
Followers431
-
Following485
-
Likes151
For those who missed the talk and want to learn more about how to use CUDA C++ to implement custom CUDA kernels training GPT-like models, watch this!
For those who missed the talk and want to learn more about how to use CUDA C++ to implement custom CUDA kernels training GPT-like models, watch this!
Cannot wait for today talk by Jake and @g_evtushenko !
Cannot wait for today talk by Jake and @g_evtushenko !
Nom nom. Amazing!
I believe this is one of the critical discussions of our time if we truly want to address diversity and inclusion in AI/ML area. Expecting under-represented community members to have access and capabilities to publish and yet be part of top schools or research labs is impossible
I believe this is one of the critical discussions of our time if we truly want to address diversity and inclusion in AI/ML area. Expecting under-represented community members to have access and capabilities to publish and yet be part of top schools or research labs is impossible
Well written article from @dylan522p and team. Covers some of opinions that I have shared privately with a few stakeholders. Clearly, CXL if they need to succeed they must significantly upgrade!
Well written article from @dylan522p and team. Covers some of opinions that I have shared privately with a few stakeholders. Clearly, CXL if they need to succeed they must significantly upgrade!
"Inference is an install base problem."
"Inference is an install base problem."
This is so true. Don't get me wrong, these days 30 under 30 or 40 under 40 mean nothing to me. Mainly because it does not matter as the real work is done by someone unrecognized.
This is so true. Don't get me wrong, these days 30 under 30 or 40 under 40 mean nothing to me. Mainly because it does not matter as the real work is done by someone unrecognized.
Legal immigrants includes Scientists (NASA, NSF, and others), extraordinary skilled workers who are entrepreneurs, innovators contributing billions of dollars to US economy each year. @elonmusk, Surprisingly many senators, congressmen from both patrties are ignorant of this!!
Legal immigrants includes Scientists (NASA, NSF, and others), extraordinary skilled workers who are entrepreneurs, innovators contributing billions of dollars to US economy each year. @elonmusk, Surprisingly many senators, congressmen from both patrties are ignorant of this!!
Prof @AndrewYNg course was my introduction to the ML world. Although I barely understood some topics back then but in retrospect the material led solid foundation based on first principles.
Prof @AndrewYNg course was my introduction to the ML world. Although I barely understood some topics back then but in retrospect the material led solid foundation based on first principles.
This piece of advice by @sama is crux of creating great team and great products. Often people don't appreciate the importance of culture and concentrate too much on technology marvel or performance.
This piece of advice by @sama is crux of creating great team and great products. Often people don't appreciate the importance of culture and concentrate too much on technology marvel or performance.
I have worked with @socrates1024, and he is a fantastic collaborator, a great researcher, and a wonderful advisor (apart from being one of the smartest people that I have had the pleasure of working with). Please apply!
I have worked with @socrates1024, and he is a fantastic collaborator, a great researcher, and a wonderful advisor (apart from being one of the smartest people that I have had the pleasure of working with). Please apply!
Swayam Singh @_s_w_a_y_a_m_
284 Followers 780 Following देखा एक ख्वाब तो ये सिलसिले हुए ✨ ML and Stuff!!Akshat @humancalico
138 Followers 1K FollowingDivya Rani @heydivyaa
744 Followers 1K Following MTS 3 @vmware | Generation Google Scholar’19 | GSoC'2019 with CERN-HSF | Outreachy'17 @opendatakitJonah Turner @drexalt
331 Followers 949 Following grinding ml, current master's student 🇫🇷 e/acc - gpu kernels - computer visionHamid @hamiddaghighEng
0 Followers 51 FollowingJoe Mathai @joemathai_
78 Followers 1K FollowingFelix @felix_red_panda
3K Followers 2K Following CS Student, speech synthesis and LLM nerd, DMs openDepay @haziz2170
10 Followers 302 FollowingAzoth @Azoth42
54 Followers 409 FollowingKevin Delnoye @KevinDelnoye
55 Followers 1K FollowingInstanton @1nstanton
55 Followers 598 FollowingJames Parsloe @jamesparsloe
196 Followers 5K Following ML Engineer. Trying to increase the FLOPs I have access to. Used to make computers talk at Spotify/Sonantic.adddddd @anandnew38
334 Followers 5K FollowingKartikay @Kartikayb77
1K Followers 283 Following Firmware engineer handling enterprise servers | logging my consciousness for ai to train on | optimist | personal blog : https://t.co/QUuAYNEnWrAaditya ; @Aaditya26082004
560 Followers 7K Following CS'26 • Machine Learning • Open-Source • Web Dev. • Algorithms • Jai Shree Krishna 🦚🪈Adela Abraham @snwl47uusx2jf
4 Followers 200 Following Startups | Fintech | E-commerce | Operator and investorPradyumna @PradyuPrasad
8K Followers 821 Following Unpaid intern for @PradyuPrasad NUS CS '27 alt: @altdyumna.s ! n h @ G @inpursuitofvoid
312 Followers 4K Following ⚠️#UnderConstruction!! *PERSONAL VIEWS ONLY, RT/LIKES ≠ ENDORSEMENT* Tech, Data, Finance, Stats, Cloud, Python, Chess, Philosophy... Peace! Here to learn..Shriram Balaji @shrirambalaji
1K Followers 5K Following engineering at @microsoft • tinkering with things on the web, typescript, rust • ⚭ @swetha__ramanVishal @eigenVectorizer
112 Followers 603 Following Mathematical Modelling, Applied Math, Cloud Architecture, ProgrammingKapil Dutta @duttakapil
998 Followers 5K Following novice generalist, on a journey to master everythingAlfred S Nainggolan @_alfredsn
2 Followers 145 FollowingAditya Maurya @AdityaM39736852
38 Followers 100 Following -Interest in deep diving into computer science. -2023 CS graduate from MNNIT Allahabad.Sirius @SiriusZeno
35 Followers 274 FollowingNishant @itsnishant14
98 Followers 788 Following Machine learning engineer @sharechatapp. @iitroorkee '22.Mustafa Zaki Assagaf @mustafasegf
4K Followers 1K Following I code Rust 🦀, Nix ❄️, and Typescript 📄. PL theory, Functional programming, and system programming nerd | CSUI 19 | Indonesian 🇮🇩Brock @BrockMcKean
3K Followers 3K Following Building @ThrivanaApp @DialetheiaApp Summoning @mettageist | e/accSurykant Rodi @SurykantRodi
129 Followers 1K FollowingDr ™©®™ 🚀 @ste_alth
51 Followers 3K FollowingPUNEET MATHUR @pmathur_cs
211 Followers 678 Following CS Ph.D. @ Gamma Lab, UMD | AI Research Intern @MetaAI | Ex- @Adobe, Dataminr https://t.co/OOJysMLncxRithsek Ngem @rithsek
76 Followers 571 Following A function of man is to live, not to exist. Data@UofOklahoma, ML@RMSDivya Shah @divya05101998
130 Followers 5K FollowingAchyut Paudel @achyutpdl
76 Followers 475 Followingspaghettski @spaghettski
107 Followers 306 FollowingPrhp1 @mojrad24
17 Followers 4K Following not a bot, Just someone with a strong thirst for knowledgeJames Kramer @theJamesKramer
1K Followers 1K Following 🇬🇧 | 28 | Father | 🔈 Speaker Head of HR @uniaptio | Prev @figma , @ScyllaDB , @hedera Building on @Blast_L2 #BitcoinSanthi R @rsanthi20
15 Followers 255 FollowingSwaggyP @PrimetimePotas
42 Followers 560 FollowingHigh Yield @highyieldYT
3K Followers 91 Following Tech Youtuber. Analyzing hardware and chips of all sizes. Everything silicon.Georgy Evtushenko @g_evtushenko
577 Followers 267 Following Member of CUDA C++ Core Libraries team @nvidia. @cuda_community organizer. Opinions are my own.Mayank Agrawal @_magrawal
505 Followers 255 Following building @roundtabledotai prev: comp cog neuro phd @ princeton, xc + t&f @ swarthmoreRoundtable @RoundtableDotAI
115 Followers 10 Following Roundtable is the modern survey fraud and bot detection platform for market research agencies and panelsmatt hardy @mdahardy
556 Followers 561 Following building https://t.co/ZgtDsnaPIA (ycs23). prev phd @princeton, undergrad @uoftVoicenotes @voicenotesai
381 Followers 109 Following “the smartest note-taker you’ll ever use” - many people. Available on the Web, iOS and Android. Coming soon to smart watches that you otherwise barely use.Deedy @deedydas
69K Followers 4K Following Investing at @MenloVentures. Formerly founding team @glean, @Google Search. @Cornell CS. Tweets about tech, immigration, India, fitness and search.The Aarthi and Sriram.. @aarthisrirampod
10K Followers 3 Following Aarthi Ramamurthy (@aarthir) and Sriram Krishnan (@sriramk) host conversations with people who have made it from being outsiders to now insiders.Nima @badnima
2K Followers 2K Followingpedma @pedma7
37K Followers 259 Following Sharing Systematic Trading Strategies Research | Building a one-person multi-strategy trading business I Systematic momentum trader | Not Financial AdviceCognition @cognition_labs
124K Followers 19 Following Makers of Devin, the first AI software engineer. We are an applied AI lab focused on reasoning, and code is just the beginning. Join us: https://t.co/tpfZwEwGiqSong Han @songhan_mit
6K Followers 145 Following Assoc. Prof. @MIT, Distinguished Scientist @NVIDIA, cofounder of DeePhi (now part of AMD) and OmniML (now part of NVIDIA). PhD @Stanford. Efficient AI computingYi Tay @YiTayML
29K Followers 97 Following chief scientist / cofounder @RekaAILabs 🫠 past: research scientist @google brain 🤯 currently learning to be a dad 🍼General Catalyst @generalcatalyst
79K Followers 2K Following We invest in powerful, positive change that endures.Yao Fu @Francis_YAO_
14K Followers 2K Following PhD @EdinburghNLP on LLMs and Machine Reasoning. Ex. @Columbia @PKU1898 @MITIBMLab @allen_ai AGI has yet to come, so keep runningMonica Lam @MonicaSLam
2K Followers 42 Following Professor, Computer Science Department at Stanford University.Yijia Shao @EchoShao8899
2K Followers 281 Following CS Ph.D. student @StanfordNLP. Previous: undergraduate @PKU1898.Christos Kozyrakis @kozyraki
541 Followers 10 Following Christos is an Associate Professor of Electrical Engineering & Computer Science at Stanford.Steve Downey @sdowney
1K Followers 1K Following Software engineer at Bloomberg LP Views are my own he/him Parody of a real software engineer and grown up. @[email protected]Aniket Rege @wregss
561 Followers 322 Following PhD @WisconsinCS | MS at @uwcse/@RAIVNLab ML, Computer Vision | Ex @nvidiaai and @samsungresearch. He/HimMark Saroufim @marksaroufim
9K Followers 658 Following @pytorch dev nowadays interested in performance https://t.co/6KJ328JUwvAndreas Köpf @neurosp1ke
5K Followers 453 Following Exploring ways to algorithmically model our world.Jeremy Howard @jeremyphoward
222K Followers 5K Following 🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Hon Professor: @UQSchoolITEE ; Digital Fellow: @StanfordMinn @minney_cat
13K Followers 903 Following building @lighthousehq_ to bet on America and help 10x more people apply for an O1 visa | prev cofounder @plymouthstreet, @bloombergbeta, @beondeck 🇺🇸🇰🇷Prashant Nair (praś�.. @Prashxnt_Nair
509 Followers 312 Following Assistant Professor, The University of British Columbia (UBC) 🇨🇦 | Computer Systems and Architecture — Tweets may not represent the opinions of my employer.Brian Beeler @BMBeeler
1K Followers 169 Following Owner of @storagereview and member of @QueenCityAngelsY Combinator @ycombinator
1.3M Followers 336 Following We help founders make something people want. Subscribe to our newsletter: https://t.co/sjqjxxBeLclmsys.org @lmsysorg
39K Followers 173 Following Large Model Systems Organization. We created Vicuna and Chatbot Arena! Compare 30+ LLMs (GPT-4/Claude/Llamas) side-by-side at https://t.co/IDFeIDIOtmMichael Isaev @michael_isaev
29 Followers 60 FollowingGreg Kamradt @GregKamradt
25K Followers 726 Following Building AI + B2B products 🖥️ Content: https://t.co/kLERwNtzqi Feedback is great: https://t.co/A6mrmjCem5 Prev. @digits @salesforceAnn Bordetsky @annbordetsky
10K Followers 2K Following Partner @NEA, early stage AI, Consumer, SaaS | BoD @perplexity_AI @contra | I like building with founders 🤖🛠️ | views expressed here are my ownRon Conway @RonConway
114K Followers 75 FollowingRishabh Srivastava @rishdotblog
12K Followers 1K Following Co-Founder @DefogData (YC W23). Previously founded https://t.co/G0jJ2DvTeR. Data nerd 🤓Vinod Khosla @vkhosla
633K Followers 575 Following entrepreneurship zealot, grounded technology possibilist, believer in the power of ideas, passionate about sustainability & impactAbhinav Upadhyay @abhi9u
10K Followers 2K Following Passionate Programmer - writes about AI, Python, Compilers, Systems Programming, Unix. Subscribe to my newsletter at https://t.co/ymkZXjD6V8raulpuri.eth @TheRealRPuri
6K Followers 329 Following AI things @ OpenAI - GPT4V, GPT4, GPT3.5, Codex | past: NVIDIA - megatron, sentiment neurons | go bears 🐻killian @hellokillian
23K Followers 439 Following building a universal interface between language models and computers ● https://t.co/yJVGuC0xlDSanjeev Sanyal @sanjeevsanyal
269K Followers 494 Following Writer, economist & collector of old maps. This twitter handle belongs primarily to the writer, and only occasionally to the economist. RTs are not endorsementsKanjun 🐙🏡 @kanjun
17K Followers 487 Following understanding human & machine minds to build a creative abundant future. CEO @imbue_ai. support founders @outsetcap. co-organize https://t.co/H1aXYk96ja.The Kobeissi Letter @KobeissiLetter
510K Followers 514 Following Official X account for The Kobeissi Letter, an industry leading commentary on the global capital markets. Email us: [email protected]Bhags @bhags__
8 Followers 33 FollowingBill Dally @BillDally
681 Followers 0 Following Chief Scientist and Senior Vice President of Research at NVIDIA. Former Chair of Computer Science at Stanford University. Former Professor of CS at MIT.@msharmavikram Not a crazy amount in compute costs - but so much in "human" costs. Could focus on improving the data mix, trying out more initial learning rates, trying out different learning rate schedules etc in a single night (instead of over many many days)!
🔥llm.c update: Our single file of 2,000 ~clean lines of C/CUDA code now trains GPT-2 (124M) on GPU at speeds ~matching PyTorch (fp32, no flash attention) github.com/karpathy/llm.c… On my A100 I'm seeing 78ms/iter for llm.c and 80ms/iter for PyTorch. Keeping in mind this is fp32,…
Have you ever wanted to train LLMs in pure C without 245MB of PyTorch and 107MB of cPython? No? Well now you can! With llm.c: github.com/karpathy/llm.c To start, implements GPT-2 training on CPU/fp32 in only ~1,000 lines of clean code. It compiles and runs instantly, and exactly…
@msharmavikram @akankshanc @rzsgrt @CohereForAI @jeremyphoward Really appreciate the offer!! We do have a private discord channel in C4AI for the purpose of discussion and since the cohort has started and materials have been posted I'm not sure how smooth the migration would be. Although we can inform the members about cuda mode discord so…
thanks to @JeffDean and @SingularMattrix for their great leadership today; and @fchollet @dwarak and many others at @GoogleDeepMind for quickly charting a good and aligned path forward together. We can go back focusing on the unlimited amounts of good work ahead of us. (Jeff,…
Is Cosine-Similarity of Embeddings Really About Similarity? Netflix cautions against blindly using cosine similarity as a measure of semantic similarity between learned embeddings, as it can yield arbitrary and meaningless results. 📝arxiv.org/abs/2403.05440
@msharmavikram The probability distribution will contain the prob of all possible tokens, so you can always see the prob of the correct next token from the text being compressed. If the model hallucinates, it will have lower probability for the correct next token, therefore poor compression.
@msharmavikram Lossless, using arithmetic coding for compression. At every step they know the actual next token and they have the probability dist for the next token from llm. They feed this dist to the coder to compress. No hallucination because they know the actual next token
I'm thrilled to see my CUDA explainers already making a difference. The gang at @Mobius_Labs have just sped up their already-awesome HQQ dequantization by using CUDA -- and were kind enough to credit my tutorial! ❤️ As @karpathy said: accessibility matters.
Yes. Developers on PC GPUs are the key enablers to DC GPU success. So all the dev tools need to work flawlessly on PC GPUs. Currently this is largely true with Geforce. Radeons definitely got better these past 6 months and they are showing increased commitment to PC developer .…
@RajaXg I think AMD have screwed up in their inconsistent approach across consumer and data center cards, different feature sets limit developers to using cards that are harder to source. Production MI300 with devs using 7900XTX would make AMD much more practical.
If you liked my "Getting Started With CUDA for Python Programmers" video, then you might want to tune in this weekend for the sequel. I'll cover writing tiled kernels that use shared memory and thread sync to take advantage of the full speed of your GPU discord.gg/cudamode?event…
@msharmavikram @GoogleColab @neurosp1ke @WenmeiHwu That would be amazing - just sent you a DM.
My mom, at 60 years of age, just graduated with a Masters in Computer Science at the University of Illinois, Urbana-Champaign (with stellar grades too)! She doesn't believe you can be too old to be curious, which is a core value for me.
@msharmavikram Thank you, Vikram! I don't think that article would have turned out the way it did without your contributions. And, I think my writing improved in that process :) Happy new year!
I worked at Intel on Larrabee applications in 2007. Then I went to NVIDIA to work on ML in 2008. So I was there at both places at that time and I can say: NVIDIA's dominance didn't come from luck. It came from vision and execution. Which Intel lacked.
@msharmavikram @OpenAI There's a lot of things you can do with embeddings (clustering, classification, etc) but retrieval is the main thing we anticipated use for. You would use ada-002 to embed your query and then rank vectors using semantic similarity (cosine similarity). We also include BM25 sparse…