Tanay Mehta @serious_mehta Twitter profile | Pikagi

Pikagi

Tanay Mehta

@serious_mehta

6,653

Followers

859

Following

379

Media

5,500

Statuses

deep learner

Bath, London, Jaipur

https://t.co/kEOk0D7Sz4

Joined July 2020

Don't wanna be here? Send us removal request.

Pinned Tweet

@serious_mehta

Tanay Mehta

2 months

I have benefited (present tense) immensely from the Open Source community and the free resources on the internet throughout my learning journey, whether it be for Machine learning, Open Source contribution help, MS guidance or general help. I think it is time for me to start

7

1

58

Last Seen Profiles

@Q_So_Gp

@creaturedesiles

@chillichelli

@coralinevermore

@Alagpro

@bwantamu_R

@okJLUV

@CCA_Activities

@SURF_NL

@evligiz60125833

@tnfmk

@CancerOsons

@prismmh

@MoniqueBak79400

@LeiLandreth

@LibhoGeza

@CharleneWa71246

@usdaxrugby

@okJLUV

@bs_3key

@Rajkuma21585640

@IpsosDenmark

@hiraganawhiter

@Jessica91763941

@Reskell492

@GraceDanoski

@AmandaT09351891

@XCHA520

@BlazeUnited14U_

@rmqueen_VT

@PatriciaGi61978

@Evan_gugu

@XCHA520

@PaytnMonticelli

@el_tren_de

@Near_Mates

@serious_mehta

Tanay Mehta

9 months

@stats_feed 1. Parts of China were colonies of several European nations 2. Nepal and Bhutan, though not directly controlled by Britain, they were British Protectorates 3. Liberia was an American colony (so technically not Europe but they were colonised) 4. Mongolia was under the control of

85

33

1K

@serious_mehta

Tanay Mehta

4 months

I don't wanna get a PhD but wanna work as a Machine Learning Engineer. Dilemma of the Century

81

48

850

@serious_mehta

Tanay Mehta

2 years

I never for once imagined that I'll be writing these lines on a piece of paper but apparently Indian CS degree had other plans in mind.

Tweet media one

48

24

570

@serious_mehta

Tanay Mehta

2 years

Indian colleges should also have Open-Source contribution as an option for the Final Year major project. Like say, the rules can be that the Open-Source project should have more than 100 stars and the contribution should be of more than 100 lines of code and should be merged.

37

47

551

@serious_mehta

Tanay Mehta

7 months

@UpdatingOnRome Oh I miss those times when the Germanic people fought with the Romans over the control of Moscow

3

2

539

@serious_mehta

Tanay Mehta

1 year

@O42nl Imagine multiplying matrices like hell, left and right, only for the results to start with, “As a Large Language Model trained by OpenAI, I cannot…”

10

12

501

@serious_mehta

Tanay Mehta

2 years

Is there a Linux distribution that comes pre-installed with all Data Science and Machine Learning libraries and Conda and everything exactly in place? The aim would be to resume working on your ML work within 30 minutes of installing the OS. I would honestly pay for that.

64

40

486

@serious_mehta

Tanay Mehta

4 months

It’s currently 1.53 AM and I am reading the RWKV paper at Abu Dhabi International Airport waiting for my connecting flight. The grind will not stop, not even at midnight.

Tweet media one

22

7

459

@serious_mehta

Tanay Mehta

2 years

Did I just get a 10 GPA in my final semester of CS??? 🤯🤯

Tweet media one

52

9

439

@serious_mehta

Tanay Mehta

6 months

Should I make a YouTube channel where I share, among other things, how to contribute to Open Source Machine Learning projects + other Machine Learning and NLP knowledge?

49

9

380

@serious_mehta

Tanay Mehta

9 months

@stats_feed source: trust me bro

12

0

367

@serious_mehta

Tanay Mehta

2 years

Practical ML Idea: An ML model that can generate a good “Subject” based on the email's body. I don't know why, but it's always a pain for me to write good email subjects.

25

21

353

@serious_mehta

Tanay Mehta

11 months

@Gerashchenko_en Buy 1 - Get 1 FREE

2

9

346

@serious_mehta

Tanay Mehta

2 years

If you, like me, get turned on by heavily Optimized and memory-efficient PyTorch code then you are in luck because I am in process of writing a thread on how to write super-duper efficient PyTorch code that can run pretty large models on Colab & Kaggle Kernels 🙃

15

18

346

@serious_mehta

Tanay Mehta

10 months

@HSajwanization No it’s not?! Our talent, our money, our engineering. Period.

1

1

291

@serious_mehta

Tanay Mehta

2 years

This morning, I became a Kaggle Grandmaster ✨🥺 Looking forward to continuing my work on making informative training and inference kernels in different competitions and datasets and helping the community 🚀 Also, I don't think someone has ever become a GM at #69th rank 😅

Tweet media one

26

4

299

@serious_mehta

Tanay Mehta

2 years

Github Co-pilot for Jupyter Notebooks, please!!

13

17

288

@serious_mehta

Tanay Mehta

3 years

ML Research Idea: Model that can annotate and explain ML papers.

9

27

273

@serious_mehta

Tanay Mehta

2 years

I turned down an offer from a company in ML and AI space recently. Why? Because despite being Open Source itself, one of the conditions was that I will not be allowed to do Open Source contributions anywhere else while I am an employee. Seriously? Not trading my freedom!

17

10

265

@serious_mehta

Tanay Mehta

8 months

@TheAnkurTyagi No amount of experience justifies a 3 LPA job in India in 2023. A roadside pani puri seller earns way more than that. For the love of god please don’t justify these low paying exploitative jobs and toxic work culture as “great potential to grow”

6

4

251

@serious_mehta

Tanay Mehta

2 years

Let's make 2022 the year for Open Source Contributions 🚀

10

17

248

@serious_mehta

Tanay Mehta

2 years

Kept this under the wraps for some time but saying this out loud now: I got acceptance for a Master in CS from a German TU last week 🥺🌟

46

2

250

@serious_mehta

Tanay Mehta

1 year

Here you go! I have published my GPT training notebook on kaggle. It features a *new* way of Data loading using PyTorch data loaders and is powered by @LightningAI for quick, clean and elegant model training along with @weights_biases logging!

PyTorch Lightning GPT Pre-training + W&B ⚡️

Explore and run machine learning code with Kaggle Notebooks | Using data from SciFi Stories Text Corpus

4

36

243

@serious_mehta

Tanay Mehta

2 years

If you are someone who wants to start reading Research papers in ML and looking for some motivation to start, then I have something for you! Here's the Notion database of the papers I've read with their summaries. You can follow my learning journey!

Tweet card media

Research Paper Tracker | Notion

Here’s a list of research papers in Deep Learning and Reinforcement learning domain that I’ve read and the ones I am currently reading.

free-bird77.notion.site

9

45

239

@serious_mehta

Tanay Mehta

5 months

When Machine Learning meets esoteric Hinduism

@c0mputerist

omnom

5 months

Tweet media one

27

225

1K

2

13

234

@serious_mehta

Tanay Mehta

2 years

Super happy to announce that I just became an Open Source contributor @huggingface transformers🤗 I have added the Poolformer model (from paper: "Metaformer is actually all you need for Vision") by Sea AI Labs. Example snippet down below 👇

Tweet media one

16

20

240

@serious_mehta

Tanay Mehta

3 months

Now I have become SGD, accelerator of negative gradient down the slope

Tweet media one

10

12

222

@serious_mehta

Tanay Mehta

2 years

I recently switched to a MacBook Pro and I am already so surprised by its battery life!? 5 Chrome tabs, including Youtube playing music for about an hour, and my battery is down from 100% to 98%?? What sorcery is it?

26

5

222

@serious_mehta

Tanay Mehta

4 months

Got my acceptance at Oxford Machine Learning Summer School 2024, in-person mode! Can’t wait to meet you all amazing people this Summer at Oxford 🚀

Tweet media one

14

2

214

@serious_mehta

Tanay Mehta

2 years

Just found out that @kaggle has actually included one of my notebooks that used Jax + @huggingface transformers + @weights_biases tracking for Sentiment Classification as one of the example notebooks for the upcoming Google Open-Source Expert Prize!

14

7

207

@serious_mehta

Tanay Mehta

11 months

@Tendar Civil war it is then?

12

1

192

@serious_mehta

Tanay Mehta

1 year

When you have no clue what LLMs actually are and what we mean by parameters but you still tweet —

@danapke

Daniel Apke

1 year

@nhutter28 Here is a scary look at where we are vs the knowledge GPT 4 will have.

Tweet media one

143

464

3K

13

8

186

@serious_mehta

Tanay Mehta

2 years

-> Got out of post-COVID depression -> Became a Kaggle Master and now in top-100 in the category -> Bagged an ML Internship from NVIDIA -> Improved my Grade point average by .4 points in one go -> Got featured in my college's annual tech newsletter for the 3rd consecutive time!

14

6

183

@serious_mehta

Tanay Mehta

11 months

@latestinspace Just imagine how big a star that once was if it "collapsed" and can still fit 30 Billion suns!

8

4

183

@serious_mehta

Tanay Mehta

3 years

Finally, I am a Research Intern!

0

0

177

@serious_mehta

Tanay Mehta

2 years

Got offered an ML Intern position at a non-Indian company. How much are they gonna pay me a month for 40 hours per week? $225 Yup. My (remarkably excellent) plumbing guy makes more than that 🤯 Workplace exploitation is real folks.

15

5

173

@serious_mehta

Tanay Mehta

8 months

Officer on UK Border Immigration yesterday: “…So you have worked in AI right? Do you think we can deploy AI on Blockchain?” Me: *proceeds to explain him how Blockchain, GPT and generally LLMs work for 20 straight mins* Him: “😳 Have a fantastic education, sir! *stamped*” did

11

2

172

@serious_mehta

Tanay Mehta

2 months

One of those days

Tweet media one

5

1

170

@serious_mehta

Tanay Mehta

2 months

Announcing the LLM Adventures Notebook series on @kaggle , where I will be making notebooks on various interesting use-cases of LLMs and RAG pipelines using Open LLMs and datasets from Kaggle ✨ Check it out:

Tweet media one

1

28

168

@serious_mehta

Tanay Mehta

6 months

I remember learning CUDA C like 2 years ago and then never using it after my internship at NVIDIA was done. Biggest mistake. Now I am going to learn it again (because ✨TRITON✨) and it feels like starting from scratch :(

6

1

160

@serious_mehta

Tanay Mehta

1 year

So technically I have worked at FAANG?!

@mbdailyshow

Morning Brew Daily

1 year

How it started How it's going

Tweet media one

Tweet media two

111

809

12K

2

1

155

@serious_mehta

Tanay Mehta

2 years

I've published yet another PyTorch training notebook in the AI4Code competition on #Kaggle . This one's using Microsoft's CodeBERT model (thanks to @huggingface 🤗). It includes optimizations, a trainer module and @weights_biases logging & exp. tracking 🚀

Tweet card media

AI4Code PyTorch Training - 🤗 CodeBERT + W&B 📉

Explore and run machine learning code with Kaggle Notebooks | Using data from Google AI4Code – Understand Code in Python Notebooks

1

17

150

@serious_mehta

Tanay Mehta

6 months

The notebook that followed my talk in London is now out! If you want to understand, code and train your own GPT, take a look at it! Modify it, pull it apart and change it as you see fit 🚀 The data loading part is now chill thanks to @lancedb !

Tweet card media

GPT from scratch using Lightning ⚡️ and Lance 🎯

Explore and run machine learning code with Kaggle Notebooks | Using data from No attached data sources

2

35

150

@serious_mehta

Tanay Mehta

2 years

Update on an experiment I did a month ago: I asked here on twitter if I should add a separate "Open Source Contributions" section on my CV, which I did. Of the 3 companies I had applied to with this new CV, I received an initial interview call from 2. Success! cc @amuldotexe

5

4

145

@serious_mehta

Tanay Mehta

10 months

@Ravisutanjani I don’t know what’s worse, people doing the actual thing or the ones justifying it in this comment section.

3

6

134

@serious_mehta

Tanay Mehta

3 months

Although the video is coming soon; if you want to pre-train or fine-tune the Mamba model as a Code-completion LLM (like Github Copilot) using a Lance dataset, I have created a repository with training scripts for you 🚀 Only supports single GPU for now

Tweet card media

GitHub - tanaymeh/mamba-train: A single repo with all scripts and utils to train / fine-tune the...

A single repo with all scripts and utils to train / fine-tune the Mamba model with or without FIM - tanaymeh/mamba-train

4

22

145

@serious_mehta

Tanay Mehta

1 year

Thrilled to announce that I will be joining @tuBraunschweig as a Masters's Student in Data Science for the Summer Semester of 2023! Can't wait to move to Germany and embark on this new journey 🚀

Tweet media one

22

0

143

@serious_mehta

Tanay Mehta

3 years

Started my transition from PyTorch to Jax-Flax. Hoping to complete this transition in a few months. Also, will soon start pushing dominantly JAX-based TPU notebooks on Kaggle, details will follow!

5

6

140

@serious_mehta

Tanay Mehta

2 months

Oh Istanbul, you have my heart ❤️🇹🇷

Tweet media one

Tweet media two

Tweet media three

Tweet media four

12

5

138

@serious_mehta

Tanay Mehta

3 years

Ok hear me out: "Code Reading Groups" Just like Paper Reading groups, except we collectively read and try to make sense of big opensource projects like Tensorflow and PyTorch. I'm sure I'm not the only one who tried reading code from such a repo and couldn't understand anything

18

5

134

@serious_mehta

Tanay Mehta

2 years

My Pull Request for adding the Hinge Loss function to @DeepMind 's Optax has been merged today! Going to add many more loss functions to Optax (for all you JAX geeks out there 😉)

Tweet media one

6

3

133

@serious_mehta

Tanay Mehta

2 years

Final exams of my engineering degree are over. Tanay is a free elf now 🌚

12

0

129

@serious_mehta

Tanay Mehta

1 year

OMG! Kaggle Models is finally a thing 😍 I remember talking to @Rob13Ell a few months ago where we brainstormed a lot of interesting ways this could turn out and it's so fun seeing it live after all! Job well-done @kaggle team 👏🏼

Tweet media one

5

11

131

@serious_mehta

Tanay Mehta

15 days

life update: reinforcement learning assignment has me coding 12+ hours everyday for the last week what even is this

15

0

127

@serious_mehta

Tanay Mehta

1 year

It's official: I am now a Computer Science graduate! Time to celebrate with a slice of pizza and a well-deserved Netflix marathon🍕 But seriously, all jokes aside, I'm grateful to have made it through this program and can't wait to see what the future holds 🚀

Tweet media one

13

1

124

@serious_mehta

Tanay Mehta

1 year

Been writing CUDA kernels since 5 AM, I can't see straight anymore someone please send help

10

1

121

@serious_mehta

Tanay Mehta

1 year

What's common is that they all left India to make a better life for themselves because our terrible politics, workplace exploitation, casteism and misogny won't let them build one here.

@varinder_bansal

Varinder Bansal 🇮🇳

@varinder_bansal

1 year

What’s common??? CEO of Google CEO of Microsoft CEO of Adobe CEO of YouTube CEO of Mastercard CEO of Pepsi CEO of IBM CEO of Netapp CEO of Nokia CEO of Novartis CEO of Deloitte

5K

2K

27K

7

6

121

@serious_mehta

Tanay Mehta

2 years

The real reason why I am doing Open source contributions aggressively is so that I can have all these organizations as infinity stones xD

Tweet media one

2

2

116

@serious_mehta

Tanay Mehta

3 months

Sundays are for creating datasets using @lancedb on @LightningAI Studio so they can be released on Monday 🚀⚡️

Tweet media one

4

6

114

@serious_mehta

Tanay Mehta

6 months

The grind stoppeth not. Open Source ML 🙌🏻

Tweet media one

7

1

107

@serious_mehta

Tanay Mehta

10 months

@JohnArnoldFndtn “that’s how an RBMK reactor explodes”

3

0

100

@serious_mehta

Tanay Mehta

3 years

Writing PyTorch training pipelines is something I will *never* get bored of

6

2

98

@serious_mehta

Tanay Mehta

3 years

Twitter fam, I need suggestion for an Ubuntu based distribution! I am going to wipe Windows 11 this weekend and install an Ubuntu based distro so please suggest your personal favorites! P.S: No, I don't need to try Arch Linux, Linux Mint or Vanilla Debian, please stay away 😂

71

4

99

@serious_mehta

Tanay Mehta

2 years

Everyone! Quickly enjoy your life, GITHUB IS DOWN. I REPEAT GITHUB IS DOWN

Tweet media one

4

24

96

@serious_mehta

Tanay Mehta

3 years

Staying up all night in a hackathon, coding with people while interacting and making friends was such a nice feeling. It all feels like a long time ago. Honestly, if you are not on a video call with your team, staying up at night in an online hackathon, you are missing out!

5

6

96

@serious_mehta

Tanay Mehta

3 years

We. Need. More. Jax. Tutorials.

6

1

100

@serious_mehta

Tanay Mehta

2 years

These Product design and UI/UX people are pure artists 💯

6

11

92

@serious_mehta

Tanay Mehta

1 month

I am surprised there aren’t more modules in university degrees teaching CUDA and accelerator programming given how important it is

13

8

100

@serious_mehta

Tanay Mehta

3 years

For someone new to ML Competitions, It can be really hard to navigate through and understand previous winning solutions. It won't be a bad idea to have reading groups where everyone tried to understand top solutions every week or so. As a community, we can achieve so much more!

17

5

97

@serious_mehta

Tanay Mehta

3 years

Multimodal deep learning is pretty cool

4

3

96

@serious_mehta

Tanay Mehta

1 year

Been sitting in this one for quite a while but here it is: I will be pursuing a Masters in Data Science at the @UniofBath from September this year 🚀 Can’t wait to move to the beautiful city of Bath 🇬🇧 and meet people in ML over there! #BelongatBath

Tweet media one

21

2

92

@serious_mehta

Tanay Mehta

2 years

If my city had Halloween celebrations, I would dress up as PyTorch's infamous CUDA Out of Memory Error

8

3

93

@serious_mehta

Tanay Mehta

6 months

Here's my first custom GPT model for you all HPC geeks: Presenting ✨TritonGPT✨ Want to write Triton or CUDA but don't know how? Ask TritonGPT.

Tweet card media

ChatGPT - TritonGPT

CUDA and Triton expert with concise, accurate answers.

chat.openai.com

6

8

90

@serious_mehta

Tanay Mehta

11 months

@mbmwaudi @0xgaut @khaalidbooker I always believed it could start a coup in a certain Eastern European country

4

0

86

@serious_mehta

Tanay Mehta

8 months

This is how Data Science Postgrads attend classes

Tweet media one

2

0

86

@serious_mehta

Tanay Mehta

4 months

Realised it kinda late but once you leave home you are never back, not completely

2

4

86

@serious_mehta

Tanay Mehta

1 year

Use this layoff season to compliment your skills in ML and Algorithms. Sitting and complaining won't get you further :)

6

2

84

@serious_mehta

Tanay Mehta

2 years

From my personal experience in doing Open Source contributions to frameworks and packages in the ML space, I've found that it's more Machine Learning Software Engineering than just "Machine Learning Engineering". (1/2)

5

8

87

@serious_mehta

Tanay Mehta

3 years

ML Beginners, unite!

@kaggle

Kaggle

3 years

🚀 You're invited to our super fun, beginner-friendly learning challenge, #30daysofML . First 2 weeks: rapidly cover essential machine learning skills. Last 2 weeks: join others in an invite-only, low pressure Kaggle competition. Sign up now, starts Aug 2!

Tweet media one

25

445

1K

2

6

83

@serious_mehta

Tanay Mehta

1 month

I would much rather grind to learn how to write high performance triton kernels that saves actual $$$ rather than invert binary trees on leetcode

6

3

85

@serious_mehta

Tanay Mehta

2 years

Silicon Valley S06-E02 actually predicted the Metaverse, and few of us realized it.

10

6

84

@serious_mehta

Tanay Mehta

2 years

I am sorry but I fail to understand how an Intern, working full-time (8 hours a day, 5 days a week) and working on a project that an actual full-time engineer is working on, gets 30% of the salary of the said full-time engineer. Only difference? This person is called an "Intern"

11

2

83

@serious_mehta

Tanay Mehta

2 years

And just when the exams got over, these 2 beauties arrived 💫

Tweet media one

Tweet media two

4

1

79

@serious_mehta

Tanay Mehta

1 year

My humble request to Individuals/training institutions: Please, for the love of god, stop teaching clueless beginners about Sagemaker, Azure ML and other auto-ML tools. Sure they are helpful but that is for later. In beginning, they need to understand the basics deeply first :)

7

7

79

@serious_mehta

Tanay Mehta

2 years

Note to me: Don't install cuda-toolkit on *Ubuntu-based distros from apt-get. It'll give you an older version from like 2019 and you'll be left wondering why your packages are not installing properly.

8

4

77

@serious_mehta

Tanay Mehta

5 months

Edinburgh 🏴󠁧󠁢󠁳󠁣󠁴󠁿 is unfathomably pretty with that dark academia vibe 🕯️ Especially at New Year’s!

Tweet media one

Tweet media two

Tweet media three

Tweet media four

3

0

77

@serious_mehta

Tanay Mehta

2 years

Don't open Social media (this includes twitter and LinkedIn) when you are having a bad day. Just don't.

5

4

77

@serious_mehta

Tanay Mehta

2 years

Published a new notebook in AI4Code Competition on #Kaggle . This one shows how you can train a BERT-Large model on a bigger subset of data and not run out of GPU memory thanks to several optimisations ⚙️ All this with @weights_biases logging support📊🚀

Tweet card media

AI4Code Training BERT-Large in PyTorch - W&B + AMP

Explore and run machine learning code with Kaggle Notebooks | Using data from Google AI4Code – Understand Code in Python Notebooks

0

7

78

@serious_mehta

Tanay Mehta

2 years

Sheldon Cooper from Big Bang Theory has inspired me to eventually do a PhD and there is no going back now

6

3

75

@serious_mehta

Tanay Mehta

2 months

So excited that my master's Dissertation will be basically training many LLMs to study their behaviour in depth 🔥 The next few months are going to teach me a ton about distributed model training and tracking and I couldn't be more excited 🚀

6

1

76

@serious_mehta

Tanay Mehta

3 years

I was running some PyTorch training scripts today and found out that using AMP and autocast takes 7 hours an epoch and the same epoch can be done in 1.5 hours without it. Moral of the story is: Over-optimisation and over engineering always isn't necessarily a good thing.

6

3

73

@serious_mehta

Tanay Mehta

1 year

See who got accepted into the ML x Health track at the Oxford ML Summer School 🎉🚀 I guess I'll be spending my summer training machines to become the next Dr House 🤖💉

Tweet media one

3

0

72

@serious_mehta

Tanay Mehta

3 months

Like how GitHub Co-pilot works? Want to train your own code completion LLM that infills code 🎯 I got you! Checkout this @LightningAI Studio that allows you to train Mamba with over 95% GPU utilisation and minimum CPU overhead thanks to @lancedb 🚀

Tweet card media

Train Mamba for Code infilling with Lance 🚀⚡️ - a Lightning Studio by tanaymeh

Learn how to train the groundbreaking Mamba Model for Code Infilling using Lance Dataset!

1

10

74

@serious_mehta

Tanay Mehta

2 years

Small question for Open Source twitter: If you implement a feature or do some big and meaningful contribution to an Open Source project, should you include it under "Projects" in your CV/Resume? If not, then where should you include it in your CV/Resume?

15

0

72

@serious_mehta

Tanay Mehta

4 months

@dk21 A lot of roles I am seeing (and applying at) for MLE positions (non research btw) list PhD in Minimum qualifications

10

1

74

@serious_mehta

Tanay Mehta

1 year

I bought a Macbook Pro, to not have FOMO, among other reasons. And here I am, months later, missing Linux every single day :(

8

2

73

@serious_mehta

Tanay Mehta

2 years

If you are looking for resources to learn Attention (and are a reading person like me), these 2 might just be all you need to get started: 1. 2.

0

7

73

@serious_mehta

Tanay Mehta

1 year

A little relic I found today, dates back to my Sophomore year of Engineering when I had to write differences between Python 2.x and 3.x in a lab file as part of my coursework. How far have we all come! The Indian education system, especially engineering is just 😵

Tweet media one

6

2

71

@serious_mehta

Tanay Mehta

3 years

Got a bronze medal in CLRP Competition last night which has now made me a @kaggle Competitions Expert! I am finally a 4x on Kaggle ✨

Tweet media one

9

1

71

@serious_mehta

Tanay Mehta

2 years

Contributing to Open Source ML is such a vibe, it's almost addicting

3

1

70