Swayam Singh
@swayam_agi
Followers
481
Following
848
Media
390
Statuses
1,998
ML Research @Microsoft देखा एक ख्वाब तो ये सिलसिले हुए ✨
living in the moment
Joined April 2021
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
jungkook
• 890346 Tweets
Arsenal
• 589713 Tweets
#خلصوا_صفقات_الهلال2
• 328271 Tweets
#HAPPYJKDAY
• 308195 Tweets
#StandingNextToJK
• 238079 Tweets
#NeverLetJKGo
• 233315 Tweets
#GlobalPopStarJKDay
• 228980 Tweets
#พรชีวันตอนจบ
• 191439 Tweets
Rice
• 157563 Tweets
Brighton
• 105382 Tweets
Ticketmaster
• 102270 Tweets
#SixTONESANN
• 83738 Tweets
ポムの樹
• 40039 Tweets
Joao Pedro
• 36413 Tweets
#상병_전정국_생일축하해
• 36237 Tweets
#Venue101
• 28451 Tweets
McLaren
• 24408 Tweets
すいちゃん
• 21406 Tweets
ALL ROUNDER YUJIN DAY
• 20322 Tweets
Veltman
• 16708 Tweets
エネルギー補給
• 15183 Tweets
Joe Root
• 12748 Tweets
Chris Kavanagh
• 12642 Tweets
Cubarsi
• 11810 Tweets
Clemson
• 11478 Tweets
ブライトン
• 11095 Tweets
バトルシップ
• 11003 Tweets
#SHUNTO_HBD_21st
• 10311 Tweets
飛び降り
• 10039 Tweets
Pinned Tweet
Career Updates in a nutshell 🥂
1️⃣ Joining
@Microsoft
as
@MSFTResearch
RF
2️⃣ Got an offer from Quansight Labs to work on
@PyData
ecosystem
3️⃣ Became
@kaggle
Competition Expert!
4️⃣ Graduating with a BTech in CS next month!
21
2
186
While reading blogs from
@freeCodeCamp
's newsletter, I found a very interesting article. With the help of 'colabcode' package by
@abhi1thakur
, you can run VScode on Google Colab, isn't it cool.
Read full article here
#100DaysOfCode
#MachineLearning
5
69
319
Learnt a very interesting concept from
@abhi1thakur
walkthrough video for
@kaggle
30 Days ML challenge D-12, P-1.
Use model for predicting missing values in columns rather than imputing or dropping them.
It worked so well, from 1556 to 464 rank🍁
#100DaysOfCode
#MachineLearning
4
19
131
Boom!
Just received my certificate of excellence for
@huggingface
's DRL course! Huge thanks to
@ThomasSimonini
and
@huggingface
team for the mind-blowing journey through the world of RL.
It's my first time trying RL and I loved it
Can't wait to apply these skills and build more
9
6
105
@mrsiipa
Using (SIMD) <immintrin.h> you can do it in 10 micro seconds (168x times faster) 🔥
Here's a minimal example of doing 10,000 additions
Requires x86-64 arch and don't forget to use extra "-mavx" compiler flag (to enforce AVX optimization)
4
1
49
My life after chatGPT:
- I attach cover letter to all the internship applications
- I reply formally to LinkedIn messages
- I submit essays asked in applications.
Thanks chatGPT :)
0
1
39
✅ First week completed at
@MSFTResearch
Setting up GCR, compute services, Azure, requesting access to here & there & everywhere, IT support calls, Meeting people, catching vibes, etc etc
$ git RESEARCH init from next Monday 🧘🏻
till then
$ git add 'DPO.bin' 'QP/'
1
1
37
So last Saturday night I gave my first online talk on "Mamba: Zero to Hero" at the Natural Language Processing group of
@CohereForAI
✨
we broadly went through 5 research works in the sessions including Mamba, S4, etc.
Thanks to
@krypticmouse
for having me🙌🏻
3
3
31
Recently I started learning core NLP from Stanford's CS224n but with a different approach.
I decided to not just watch lectures or read notes but also go a bit more hardcore side to read and implement the Research Papers from scratch.
And the results are pretty obvious 👇
3
1
29
I just know that nobody is going to read my math-heavy blogs, but just for a moment, take a look at how beautiful this is ❤️
Post:
0
1
27
Career Update:
Joined Scaler (InterviewBit) as Data Science Intern.
Cheers 🥂!!
6
0
28
A few days back
@ehmorris
released a game (it's awesome xD).
I build a custom RL environment for it to train a PPO model and see if it can make a landing.
Luckily training it for 1 million steps (~14 hrs), it's able to make a (meh..) type of landing
5
3
24
Wow!! we just hit a stellar 150 ⭐️ milestone! 🎉
Thanks, everyone 🙌🏻
A whole lot of open-source love❤️to you all 🍻
+ Stay tuned, will public another cool repository soon 🎉
cc:
@navaneethakbh
@ParthGupta_11
@Ihtreeek
@CreworkHQ
#opensource
1
2
22
Day 102+103 :
#66daysofdata
- building habit of solving 5 leetcode questions daily.
- finally passed the week 1 assignment of Convonets of DL specialization (failed twice 🥲)
#100DaysOfCode
3
3
21
A moment of peace for those
@PyTorch
users who forget to do 👇
optimizer[dot]zero_grad()
While training model.
#100DaysOfCode
#66daysofdata
1
5
20
Looks delicious 🍽️! After reading tons of theoretical aspects around diffusion.
Saving it for the WEEKEND
Our tutorial on diffusion & flows is out! We made every effort to simplify the math, while still being correct. Hope you enjoy! (Link below -- it's long but is split into 5 mostly-self-contained chapters).
lots of fun working with
@ArwenBradley
@oh_that_hat
@advani_madhu
on this
27
275
1K
0
1
21
🎉For AI + Music fans
👀Here's a sneak peek of what I've been working on for the last few days. Listen to this sample soundtrack from "My Heart Will Go On", created entirely using
@SoundScapeHQ_
🎧Headphones recommended🎧
Stick around—there's a treat for you at the end! 🥂🧵
7
4
21
Can we go more brr.....? How about switching to GPU?
To avoid memory & kernel overhead impacts, increasing size from 10k to 100 Million 🤯 and page-locked memory for quick data transfers
The results 🎉: CUDA wins 🚀
⚪️ ~5x times faster than vanilla
⚪️ ~3x times faster than AVX
@mrsiipa
Using (SIMD) <immintrin.h> you can do it in 10 micro seconds (168x times faster) 🔥
Here's a minimal example of doing 10,000 additions
Requires x86-64 arch and don't forget to use extra "-mavx" compiler flag (to enforce AVX optimization)
4
1
49
2
0
21
It's freaking interesting (and a sense of responsibility) that a project I finished in 2022 suddenly started getting a hell lot of attention in 2024 by the industry brands.
The Demand is high and we'll fulfill it for sure.
Launching 2.0 Soon!!
3
0
21
Day 101 :
#100DaysOfCode
- completed chapter 4 of DL with PyTorch ;
learnt about time series transformation, text encoding and embedding, etc
*fun-fact: It took 6 years for Kepler for fitting an ellipse (model) to some space-info data 👇
#66daysofdata
1
4
21
🚀 Exciting news! "Clothes Virtual Try On" just crossed the 100 stars milestone on GitHub! 🌟
What started as a small idea in
@CreworkHQ
's "Build With Us" cohort has grown beyond our wildest dreams.
Repo:
#OpenSource
#GitHub
3
2
20
Day 105:
- solved 5 questions on leetcode of Binary search ( if question labelled binary search still think of better algo)
- learnt some convolutional NN architectures like ResNet, InceptionNet, mobile net, etc. And techniques to work with CNNs
#100DaysOfCode
#MachineLearning
0
9
19
In the 2nd half of 3rd year, I made this
@NotionHQ
page to help me monitor my career journey, setting goals and deadlines, monitor project progress, etc.
From someone who did not have any proper exposure and guidance, I Must say it was worth it.
3
0
19
Day 104:
- solved 5 leetcode questions on binary search
- learnt theory of week 2 of DLS: C4
#100DaysOfCode
0
7
16
Low-Level Parallel programming & system architecture after college are so fun!
College used to bore me with COA and more 🫤
What hurts is not having a dedicated CUDA machine (will get one by next month, FOMO ⬆️)
3
0
19
It was never about Placements or Money or Machine Learning or Research or Anything.
It was always about the melody and rhythm that kept my soul dancing. Amidst the chaos of ambition, I found harmony in the simple notes of life's symphony 🎶✨
3
1
18
just gave
@Microsoft
's engage test and did all the questions in 10 minutes.
HOPING FOR THE BEST :)
4
1
17
Maybe I'm not aware of this, but is it true that cracking YC startups is a big thing?
Last week itself 2 founders reached out to me from my GitHub.
2
1
17
Check out our latest work which breaks new ground in the world of Code LLMs.
1️⃣4TB permissively licensed pretraining & instruct-tune data
2️⃣Benchmark for Code LLM generalization
3️⃣Octo Duo: 2 best permissive Code LLMs
Paper:
Demo:
How to instruction tune Code LLMs w/o
#GPT4
data? Releasing
🐙🤖OctoCoder & OctoGeeX: 46.2 on HumanEval🌟SoTA🌟of commercial LLMs
🐙📚CommitPack: 4TB of Git Commits
🐙🎒HumanEvalPack: HumanEval extended to 3 tasks & 6 lang
📜
💻
1/9
7
88
292
3
2
17
Okay so I got to know my 1st year results and somehow (I don't know how and why) but I scored 9.08 CGPA.
Now I want respect 😎
#Respect
5
1
15
Day 85+86 :
#66daysofdata
Working on 3rd article and learning C++ concepts in depth.
Can't able to work properly these days, due to some issues. Hoping to get back from tomorrow
#100DaysOfCode
0
4
15
Day 52 :
#66daysofdata
- refreshing the eigenvalue part from mathematics for machine learning book.
- completed till 'pretrained networks' of Deep Learning with PyTorch book.
#100DaysOfCode
Fun fact 👇
1
4
13
Throwback to when I was preparing the presentation for my "Mamba: Zero to Hero" talk for the NLP community of
@CohereForAI
with a serene sunset as companion 😌🌄
PS: What a wonderful world!!
0
0
15
A nice, detailed and beginner friendly explanation of Decision trees by
@abhi1thakur
#100DaysOfCode
#MachineLearning
0
9
14
Just noticed an interesting pattern 😂
People see my tweets here and then visit my LinkedIn and we connect as:
"Hey saw your tweet, awesome work man, let's connect"
BTW I'm more active here && reply DMs too 🙌🏻
0
0
16
Why is DARK the best series? 👇
Time travelling, parallel worlds, quantum entanglement, bootstrap paradox, Fermi's paradox, Gothe's dualism, Higgs-Boson, String Theory, Wormholes and butterfly effect. (If I didn't miss any)
Dark is the fantasy for creatures like me 😌.
1
2
15
Reached the 1st milestone,
last 4 months and 2 amazing research works, what a nice way to end 2023
And with that, I'm off home 🏠
1
0
15
@neembu_paani31
Credit to the respective creator xD
Jokes aside man, let me know if you got any JD at MS, will check if I can help via some friends
1
2
48
The recent LLM Science Exam competition on
@kaggle
was crazy, I mean running 70B model on a single Colab/Kaggle GPU + RAG + Openbook🔥
Waiting for the writeups
0
3
15
Went college after 2 months, my batchmates told me about the project he was working on using Machine Learning.
Guess what, he can accomplish same thing with HTML & CSS in just 3 lines of code🙂.
First rule: never use ML on those tasks which can be done easily in other ways.
2
0
15
Hard times bring out the best in people ✨
Saw two colleagues (same position as me) go through layoffs, and their hustle since then has been amazing. Impressed with how they've upped their game and are killing it in the market
More power to everyone struggling—you've got this!
0
1
13
Today is the craziest day at work, I had to read and implement the 2020 research paper of U^2 net. This is the first time I was feeling that I'm no more an intern 😂.
Jokes apart, end of the day, made it worked + extra functionality ( hoping it's not gonna ruin the performance)
2
0
13
Just to clear the confusion:
1. Writing CUDA kernel
2. Writing efficient CUDA kernel
Are both way too different things and impacts are big, it's the difference between useful and useless.
PS: and the reason I don't appreciate ChatGPT guided "from scratch" beginner projects
2
0
14
When you combine your both love 💕 (Music + DL)
Wrote a basic LSTM model to generate music sequence.
Shift to Transformers and sell music as NFT 🤣
1
0
13
I remember when I was thinking to start with RL, I was looking at the contents of an introductory RL course and saw the term "Markov Decision process".
I got scared that 'Markov' it's somewhere related to stats/probability and I don't know that concept.
So I step back for while
3
0
13
Key is to start, and learn the stuff in the way. You may not understand the mathematical complexities but you'll be able to get the context.
Like "what is Markov Decision process" well it's just "future only depends on present not past" 🙂. We can work on maths later.
0
0
13
You have relationship insecurities and we got job insecurities 😂😂. We're not the same bro
0
0
13
Confession: Being in a core tech undergrad program I don't wanna be a SDE, etc etc at any MNC for my long term future, I like Quantum Physics and Machine Learning, I'm a research guy who builds just for fun. I like to go deep.
peer-pressure is on the verge.
GOD SAVE ME
1
0
13
Day 67 :
#66daysofdata
Started with Deep Learning specialization, completed week 1 of Neural Networks.
Fixed my schedule, need to work on physical health too 😑
#100DaysOfCode
1
2
13
@kush_soumya
OK so
1. Projects: Replace them, not satisfactory, bad description, way too traditional. If you're in ML try projects which combine software engineering + ML (Impact, tech used, MLOps) + You're a fresher so have at least 3
More 👇
1
1
13
To all the men out there,
Just never let your guard down, never ever for anybody, I'm telling y'all!!
6
0
13
Day 38 :
#66daysofdata
did lot of practical stuff, created a basic 2 layer neural network using gradient descent and perceptron algo. sharing repo for (gradient descent and 2 layer NN applied to predicting admission in UCLA)
#100DaysOfCode
@DeepLearnersBot
1
0
11
Day 89 :
#66daysofdata
Checkout the implementation of research paper on "Style Transfer using VGG-19" in PyTorch by me.
Style transfer is a technique in which we use CNNs to transfer the texture from one image to other.
#100DaysOfCode
link to notebook 👇
1
11
11
Part of journey is the end
#30daysofml
by
@kaggle
, these 30 days were really action-packed days for me. Being a total beginner, these 30 days were so productive for me.
Thanks
@abhi1thakur
with your techniques I got to achieve rank in top 3%
#100DaysOfCode
#MachineLearning
1
2
13
🥂 Life Update 🥂 :
Having interns working under you is a special feeling in itself.
0
0
13
Proud to have contributed to this project and work with the amazing BigCode community, this is the best open-source code model in the world with the support of:
✅80+ languages.
✅15B LLM with 8k context
✅Also a tech assistant
✅1 trillion tokens
✅VS Code plugin
And much more!
Introducing: 💫StarCoder
StarCoder is a 15B LLM for code with 8k context and trained only on permissive data in 80+ programming languages. It can be prompted to reach 40% pass
@1
on HumanEval and act as a Tech Assistant.
Try it here:
Release thread🧵
76
668
3K
0
1
13
Day 95+95 :
#66daysofdata
DL specialization: C2, Week 2
Working on data structures and algorithms.
#100DaysOfCode
"Machine learning is tough, in the meantime me studying ML from
@AndrewYNg
"👇
0
8
13
@Gillesvdwiele
@abhi1thakur
@kaggle
This is the function I used for predicting missing values, using average prediction of both Randomforest and Gradient boost.
earlier I used just randomforest but combining 2 ensemble methods gave much better results. Now rank with the below function is 387
1
1
13
Day 106+107
- learnt about polymorphism & exception handling in C++ in great depth. Sometimes I love the way C++ works (ignoring UBs 👀)
- attended
@CreworkHQ
's session with
@ishandutta0098
on data pipelines and the biggest takeaway that 'I need to learn a lot'
#100DaysOfCode
2
3
12
STORYTELLING is an art.
A compelling and intuitive way to convey your logical thoughts to almost everyone. It doesn't matter whether you and your audience share common knowledge or not.
Read it all if you're confused about what I mean
🧵 Thread 🧵
2
0
13
Day 56 :
#66daysofdata
- Continuing with MAchine Learning course from Courserea, completed lectures of Week 5,
tomorrow I'll derive few mathematical conclusions (just to convince myself), then implement the algorithms of assignment
#100DaysOfCode
0
10
13
Part of the journey is the end !!
Completed my internship at
@crowdanalytix_q
as Data Science Intern. Throughout my journey, I build and trained multiple models from scratch and even got the opportunity to work with some of the finest SOTAs including Stable Diffusion 🍁
7
0
13
It's been a week, I didn't work on anything at all 😕
Gotta end this vacation week now
1
0
12
When Mahadev takes care of you, your soul becomes an internal burning flame 🔥
Did 5 hours train journey standing, spend 2 nights and 1 day sleepless literally on the streets and came back 3 hours of train journey again standing
And still I can dance on his vibe
हर हर महादेव ❤️
1
0
12
Day 64 :
#66daysofdata
I'm so glad to tell you all that I completed the Machine Learning course from
@AndrewYNg
on
@coursera
within 22 days with the final grade of 95.68 %.
Thanks to
@svpino
for roadmap, 1st step done :)
Some of my takeaways 👇 🧵 (A thread)
#100DaysOfCode
2
1
11
I have been writing JavaScript code for the last 10 hours, and wrote enough Notion doc that it sometimes stops responding.
BUT I'm still not bored.
From where all this energy is coming from?
#javascript
4
0
12
Day 74 :
#66daysofdata
- Waiting for 2nd course, it'll be available on 1st Dec, so today I planned the structure of my blogs for ML series.
Going to publish an article tomorrow about the structure and what you can expect. There's a surprise in the end😁.
#100DaysOfCode
0
4
12
The best thing happened today,
Was going to work on stable diffusion at work (internship) ✨😌
0
0
12
Confusing seasoned ML engineers
This pic demonstrates that the first and last layers of a causal LM (GPT-2) share the same weights!
5
0
12
starting up again with
#100DaysOfCode
and continuing DL specialization.
Day 108: Implemented U-net, performed Object detection and semantic segmentation as part of Assignment of W3.
Day 109: Learnt and implemented face verification and recognition system as part of W4 assignment
0
2
12
The major reason I like experimenting my runs with
@weights_biases
apart of all its functionalities, is the cool namings of each run 😂.
One can literally use them as their usernames. (radiant-river-11)
1
2
11
Day 49 :
#66daysofdata
Note: I'm not good with art, but I know some Deep Learning tricks, Below is an example of today's work on implementing Style transfer using VGG-19.
Code 👇
#100DaysOfCode
0
2
11
Been a month in Bengaluru and I'm really starting to vibe with this city, it feels good...
Also realized that relationships only seem adorable & awww when I spot the street couples moment
— definitely definitely not my thing! 🙂
3
0
12
I've been reading multiple technical books these days, it's understandable that picking the right resource when just starting up can be daunting.
So this month end, I decided to share some of the resources that helped me in my journey of AI/ML.
Stay Tune :)
2
0
11
It's not burnout,
but for the first time in ages, I'm struggling to manage my time properly. A mountain of tasks and only 24 hours in a day.
Trying to find balance amidst the chaos. 🧘🏻
1
0
11
I completed my Deep Learning Specialization from
@coursera
today
Thank you
@AndrewYNg
sir, your way of simplifying complex topics is NEXT LEVEL.
#MachineLearning
#100DaysOfCode
1
5
10
Success is overrated, the real adrenaline rushes are in coming out of failures with a smirk on the face
1
0
11
This Halloween expressing my biggest fear:
CUDA: out of memory error
2
0
11
Using AI for copy pasting code to make opensource contributions can get you or organisation in big trouble.
You don't know what lisence code is being exposed by your copilot and that can later mess with the licensing of that opensource organisation
1
0
12
Resolved 14+ DMs! 😅
The notification game's been wonky,
Do let me know if in any case, I won't revert to your queries.
Giving back to the community is the last step.
I'm too highly respectful towards the people I reached out to and who genuinely guided me
1
0
10
Looks like my juniors finally figured out how to use the calendar! 😄
Got a whopping 58 Teachers' Day wishes today – guess I've officially made it to the 'semi-cool' teacher-bro status!
Thanks for the love, you rockstars! 🤘❤️
2
0
11
Recent
@DeepMind
's paper gives the highly precise mathematical overview of transformers, how they work, how they're trained and inferenced, etc.
I'll say it's a definite read and you can also practice your STORYTELLING skills on this too, making sure things make sense
2
1
11
Day 100:
#66daysofdata
4th article of *Demystifying Machine Learning* series is out. It's our first project where we are going to build a *cats and dog classifier*.
#100DaysOfCode
#MachineLearning
0
5
11