Sumanth
@sumanthd17
Followers
2,070
Following
1,465
Media
78
Statuses
1,156
PhD’ing @iitmadras @AI4Bharat , Google PhD Fellow, Past life - @GoogleAI @Mila_Quebec @IIITSC
Joined November 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
TEMU
• 430613 Tweets
最大5,000円報酬
• 226528 Tweets
領空侵犯
• 147426 Tweets
Sven
• 114661 Tweets
Arlington
• 91599 Tweets
Delpino
• 47250 Tweets
シロネン
• 46621 Tweets
Finale
• 34177 Tweets
Scout
• 27731 Tweets
Nintendo Direct
• 24296 Tweets
線状降水帯
• 23780 Tweets
Indie World
• 23769 Tweets
DUNK SMN EP4
• 22489 Tweets
PHUWIN SUMMERNIGHT EP4
• 22045 Tweets
Switchの後継機種
• 21372 Tweets
ソフトメーカーラインナップ
• 21129 Tweets
Rayo Vallecano
• 20470 Tweets
D-1 TO CHANYEOL SOLO
• 16697 Tweets
James Rodríguez
• 15186 Tweets
#NationalDogDay
• 11937 Tweets
伊藤さん
• 10258 Tweets
Pinned Tweet
🚨 Paper Alert!!
#LLMs
are increasingly used to evaluate other LLMs, impacting important development decisions. But how accurate are these Evals? We introduce FBI, a framework to the find blind spots in Evaluator LLMs.
📎 -
🔧 -
3
19
79
Happy to share that I’m one of the recipients of the
@GoogleAI
PhD Fellowship for 2023! Grateful for the support of my mentors and colleagues at
@ai4bharat
@iitmadras
who helped me get here. Congratulations to all the awardees!
And, thank you Google :)
16
6
310
Exciting news! 🎉 Just got green light on visa paperwork, & I'm thrilled to announce that I'll be joining
@GoogleAI
Mountain View for the next 4 months to work on LLMs. Looking forward to new experiences!! Bay Area friends, I'll be here until Nov, let's catch up if you're around!
14
3
291
[🚀Update] Hello Bangalore! Continuing my
@GoogleAI
stint from Mountain View, I’ll now be working from India’s Silicon Valley. Starting tomorrow I’ll join as a Research Intern
@GoogleIndia
I’ll be in Bangalore till March’24! HMU if you are around and want to talk NLP
7
3
143
After letting it all sink in, here it comes.
Happy to share that all 3 papers got accepted to ACL'23 (2 main and 1 findings). And none of this would've been possible without the wonderful peers & mentors at
@ai4bharat
Will share the papers, datasets, code & everything soon!!!
11
3
136
Walked out of the
@Google
office one last time! 4 months went by in a flash :/
Here’s a cliché picture at googleplex
0
1
110
Returning to Bangalore after an incredible time at EMNLP! Met old friends, made new ones, & had unforgettable experiences. Amidst the hustle of talks, posters, and parties, I was recognised as a decent reviewer – thanks to the ACs and authors. Hopefully I wasn’t R2 to anyone :)
4
0
108
A year ago, took the leap of faith and started my PhD.
I was part of 4 papers in different capacities - all trying to address the lack of models and resources in the Indian context
Got to travel outside the country to attend 2 confs (ACL & EMNLP) and made a good bunch of friends
3
3
107
And done with the final presentation in February. Passed my PhD qualification exam (comprehensive)
Andddddddd after almost a month hoping to open VS Code again tomorrow. Enough with the slide decks bring on some code!!!!
2
1
79
New Paper 🚨
IndicXTREME: A Multi-Task Benchmark For Evaluating Indic Languages
We introduce IndicXTREME, a diverse benchmark of 9 tasks covering 18 Indian languages. We maintain high quality by using human supervision to create all the test sets. [1/8]
4
15
69
Definitely one of my best conference experiences. Finally met my coauthors in-person, made a lot new conf buddies and ofc interesting papers all around the week. See yall at the next one!!! GG
#ACL2023
@aclmeeting
Here’s nice pictures from the Niagara Falls
3
0
67
[Post for my future self]
Today we had the opportunity to sit down for lunch with
@JeffDean
. Great mind and also a very fun person to talk to!! Thanks for your time
Of all the things, I asked about the latency numbers a programmer should know :)
1
2
63
From skipping breakfast for 2 years to now enjoying it regularly! Feels like two different worlds!!
Btw this oats and chia pudding takes like <5min to make and I feel I found a new good habit
5
1
60
Excited to share that our survey titled "A Survey of Adversarial Defences and Robustness in NLP" has been accepted to ACM Computing Surveys
Arxiv version is available at
Work w/ Shreya Goyal, Mitesh Khapra,
@ravi_iitm
2
7
58
T minus 0
Just finished an amazing first day at the office! Feels absolutely unreal to work at one of the the OG
@Google
workspaces. Capped it off with a dinner catching up with old mates
@Swarooprm7
@anuj_diwan
1
1
57
On my way to
@aclmeeting
. Let’s chat about all things Multilingual (& beyond)
I’ll be (co-)presenting 3 papers
IndicXTREME -
Naamapadam -
Vārta -
Waiting to meet so many friends!!
See you all in 🇨🇦
#ACL2023
3
2
49
So it's less than 100 days before we hit 2023 and I want to manifest to read at least 1 paper a day till the end of the year. I would be happy to get recommendations on
Primary - NMT, Massive Multilingual Training
Secondary - Speech, Calibration
New - Diffusion, Vision
[0/100]
4
2
47
I've done itttttt!! 100 days of reading. Ofc could've done some things better, but it's a start. A lot of learnings, many mistakes as well. I read decently diverse papers which helped me to develop lateral thinking* (100% would recommend)
More to come in the (Happy) New Year!!!
So it's less than 100 days before we hit 2023 and I want to manifest to read at least 1 paper a day till the end of the year. I would be happy to get recommendations on
Primary - NMT, Massive Multilingual Training
Secondary - Speech, Calibration
New - Diffusion, Vision
[0/100]
4
2
47
3
1
45
I will be at Research Week w/
@Google
today! This is my 3rd time attending the event, and for a change I'll be playing for the home team this time around!
HMU if you wanna talk NLP!
2
0
45
Can’t wait for the next *CL conference to start conversations with “Hey remember we danced together at EMNLP22 🤣🤣🤣🤣” one of the best conference experiences
#EMNLP2022
4
2
43
I'll be in Singapore🇸🇬 for
@emnlpmeeting
!
My agenda is to strike conversations on Evaluation of Language Models. English and also Multilingual
Let's complain about all things on current AutoEval and maybe find a solution? :)
Let's explore food in SG and talk NLP! DMs are open!
1
0
43
Just had an amazing day at Google Research India office! Got to dive deep into fascinating problems in speech and NLP with some brilliant minds. Thanks for having me
@nitish_gup
@partha_p_t
1
1
43
Any recommendations on how to forget 8hrs of my life?
I stand by the boys 💙
But still 😭😭😭😭
4
1
42
Until today, RoBERTa was my fav paper. I think that has changed today. Very engaging & insightful paper and most importantly BEAUTIFUL storytelling. Thanks to all the authors!!
The State and Fate of Linguistic Diversity and Inclusion in the NLP World
1
3
41
Roughly 2yrs ago we came out with Samanantar. We created the largest parallel corpus and best MT models for 11 Indian languages.
Fast-forward to now and here we are, covering all the 22 Scheduled IN languages and pushing the boundaries on data, MT quality and a new benchmark. 👇
🚨 📢 Preprint Alert
After more than a year of hard work, we are pleased to introduce IndicTrans2, the first machine translation system supporting all 22 scheduled Indic languages.
📎:
💻:
▶️:
Thread👇[1/n]
8
92
384
2
2
41
Welcome boss!
This guy is an amazing researcher and even better friend!! I guess I'll see more of you now :p
Extremely excited and humbled to announce that my appointment as an Adjunct Faculty at IIT Madras has been approved!
With this, I hope to deeply involve myself with Indian academia and work with brilliant minds, starting at AI4Bharat!
New Website:
35
6
292
0
0
40
Okay hear me out!
"Thinking" of doing a 30 min podcast (Ik cliche) where I chat with my friends doing amazing things and loving it. Would you tune in?
I planned a version where I'd cook their favorite meal and chat in person—still a bit far from that, but this is a start!
[1/3]
5
1
34
2007 T20 finals was my first every cricket match, I was 8. I saw my dad jumping with Joy 🙌
17 years later, here I am jumping with joy!
Ah WHATA GAME, you beauty
0
3
35
I told myself I’ll ride after the deadline and here I am. I’ll be riding from Cochin to Kanyakumari over the next week. Any recommendations on places to see and food to try?
Here’s me smiling at the fact that I’m riding after 2 long years!! And beautiful Alappuzha(Alleppey) town
2
1
32
Read more than 100 papers. Came up with a bunch of ideas based on the readings. The results of these are still pending but I'm glad I made the effort
Looking back, it was a decently good year (touch wood)
Let's see what 2nd year has in store
0
1
32
Absolutely agree with all the points mentioned here. I’ve personally experienced some of these things and fortune to be advised some of the best faculty. Reach out, if you are considering a PhD (in India or elsewhere), happy to answer any questions :)
To all the prospective PhDs in/from India, before you sign up for a PhD position in the US, please carefully consider if you would do better (PhDing) in India. I sketched my experience in this blog . Feel free to send me your comments 😃
19
44
291
1
3
31
Super excited to attend ACL2022 in Dublin, Ireland!
#ACL2022
@aclmeeting
. If you happen to be there, please say hello. Get in touch if you wish to discuss NMT, Multilingual NLP
#NLProc
1
5
31
The number of times I had to re-create my conda environment is just UNREAL!!
Every package wants a new torch and wants to compile on a different cuda version!!!
what's the safest version? -- Currently running on torch 2.3.0 with cuda 11.8
10
0
29
Went to a Broadway show just to “tick it off my list”
But OH BOY, MIND BLOWN!!! This is easily one of the best experiences of my life!!!
1
0
29
Another day, another effort towards multilinguality!
Keep an eye out for this lad
@Harman26Singh
New work on evaluating LLMs for generation in Indic Languages:
IndicGenBench
👉5 diverse tasks, 29 Indic languages, >100k examples.
👉Curated using human translations ensuring high quality.
👉Multi-way parallel dataset.
(1/n)
8
44
196
0
0
27
Aggressive Google Searching: Internships at Meta (Fall'24 / Summer'25)
3
0
27
Planning a visit to NYC 🗽 Oct 23-28 and got some ambitious stops on my list - NYU, Columbia, Princeton, and Google NY office!
Knock, Knock!! If any of you happen to be behind any of these doors, can you show me around? I'll get the pizza🍕
Looking forward to visiting BigApple🍎
2
3
27
Heading to Bangalore for the next few days. Among other things will be attending the Google Research Week. HMU if you are attending/want to talk about NMT/Multilingual LMs/Speech. Looking forward to meeting with folks working on similar things
5
1
24
Just had an incredible experience crossing off a major bucket list item! After seeing it countless times in movies and video games, finally got to see the Golden Gate Bridge 🌉 in-person!!
0
0
23
The application season is upon us. I got a lot of help from many students and mentors when I was doing my PhD applications last year. It’s my time to return the help. I’m happy to help students applying to MS/PhD programs in CS dept wrt SOP, colleges, anything and everything.
1
5
23
Will be attending
#EMNLP2022
@emnlpmeeting
in-person. Let’s talk if you are interested in NMT, Multilingual Models or Speech.
4
1
23
TMLR
@TmlrOrg
!!! Yayyyyyy
Congratulations to all the people who put in a lot of effort to make this happen!!! 🙌
🚨 📢 Preprint Alert
After more than a year of hard work, we are pleased to introduce IndicTrans2, the first machine translation system supporting all 22 scheduled Indic languages.
📎:
💻:
▶️:
Thread👇[1/n]
8
92
384
0
0
22
Another year, another version of the corpus!! Excited for all the possibilities this opens up. Proud to be part of this work 🚀
Led by the best
@SafiKhan2k
@PriyamMehta7
!!
🚀IndicLLMSuite Launch Announcement!🚀
We're thrilled to unveil IndicLLMSuite: A collection of data resources and tools for developing Indic LLMs.
📜 Paper:
🌐 Blog (the way forward):
💻 Resources:
(1/n)
4
51
201
0
0
22
Wanted to be an Air Force pilot, tried the exam 3 times after I started my PhD, didn't clear :( Life went a different way, but having fun now. Maybe in the next life, I'll be a pilot! 🛩️
what was your dream career as a kid and what did you end up majoring in? i wanted to be a doctor growing up and I've a degree in economics now lol
53
7
154
5
1
21
Updated all the ArXiv submissions with the camera-ready versions of ACL 2023. Here are the 3 papers with all the links; paper, code, data any summary 👇
1
0
21
Cooking is soooooo unfair. Spent 4 hours cooking for 20 people. Finished eating in <20mins 🙃
5
0
20
Started 3 hrs prior to departure, paid extra for the cab and and still I missed my flight. Why you do this to me BLR traffic 🥲🥲🥲
4
0
20
The new journal app is just amazing. Automatically clusters all my activity, photos, location based suggestions from trips. I might finally consider the digital switch from my book :/
3
0
19
Revise and Resubmit. That's what PhD students do? 🥲
0
0
19
Wake up
Refresh mail
Check for “that” email
Feel sad about not getting it (yet?)
Go on with the rest of the day
Repeat
0
1
18
It was an absolutely amazing experience to attend
#ACL2022
in-person @ Dublin. Amazing city and amazing people all around
1
0
18
Want to give a talk about your current NLP research / You'd like to listen to someone give a talk?
Nominations are now open for the Spring + Summer series at
@NLPwithFriends
NLP With Friends is opening nominations for our Spring + Summer 2023 series!👫
We want to hear from YOU — nominate people you’d like to hear from, self-nominations accepted and encouraged!
🗓Nominations close: April 5
✍️Where:
🎙Talks return on April 15!
0
8
28
0
1
18
Join here on Sunday (IST) to hear about interesting works happening in India. I will also be presenting our work “Towards Leaving No Indic Language Behind: Building Monolingual Corpora, Benchmark and Models for Indic Languages”
(1/2) We're excited to meet you at 'Data Science in India' on Sunday, 6th Aug. A series of insightful talks, panel discussion and networking opportunities awaits you at the event. If you haven't registered yet, do it now:
1
0
4
0
0
18
Excited to be in Bangkok for
@aclmeeting
#ACL2024
from Aug 10th to 14th! Looking forward to discussions on Auto Evaluation of models, multilingual models, and beyond. My DMs are open—let’s connect!
Where’s the party this time? (IYKYK)
0
0
17
15 months and finally!!! A good and welcome grind into the PhD program.
New Paper 🚨
IndicXTREME: A Multi-Task Benchmark For Evaluating Indic Languages
We introduce IndicXTREME, a diverse benchmark of 9 tasks covering 18 Indian languages. We maintain high quality by using human supervision to create all the test sets. [1/8]
4
15
69
0
0
17
Today I complete 2 years (time files) since I joined AI4Bharat, IIT Madras. This continues to be a rewarding time here with all the amazing mentors, collaborators and fellow researchers. Coincidently today is also Vijaya Dashami, starting the day with a sense of victory.
2
0
17
California, you’ve been kind. Time to bid farewell for now. Will our paths cross again in the future? Until then, stay golden!
1
0
16
3 months ago, I started this streak of 1 paper a day, currently at 85. Original goal was 100, revised is ♾
Across domains I'm curious what are NLProc people' fav papers are. Based on my interactions I'll tag few people here, add your fav paper here, if that's okay. I wanna read!
So it's less than 100 days before we hit 2023 and I want to manifest to read at least 1 paper a day till the end of the year. I would be happy to get recommendations on
Primary - NMT, Massive Multilingual Training
Secondary - Speech, Calibration
New - Diffusion, Vision
[0/100]
4
2
47
1
0
15
Finally done with one leg of the ride
📍Cochin - Alleppey - Varkala - Trivandrum - Kanyakumari
Now that the free trial is over, my new year start today! Looking forward to all the things that 2023 has to offer.
Clouds played spoil sport but such is life.
1
0
15
An unforgettable East Coast adventure! 10 days, 3 cities, and over 100km of walking exploration in NYC alone. Reconnected with old friends and made new ones – you all know who you are, and you're truly amazing. Thanks for the incredible memories! Until next time ✌️
1
0
15
Had an amazing 3 days and thoroughly enjoyed all the lectures and the interactions. Thanks to all the organisers for a successful event!!
Research Week with Google - officially a wrap!
Extremely energising to be with students and see their research curiosity! Till next time!
Special thanks to our amazing speakers, ACs, organisers and Program Chairs !
7
28
195
0
0
15
Excited to delve into the latest research and engage in thought-provoking discussions with fellow
#NLProc
enthusiasts.
Co-organising this talk series with
@zhaofeng_wu
@BrihiJ
@kayo_yin
. Stayed tuned for more updates
We’re thrilled to announce NLP with Friends will return this spring with an all-new cast of student organizers!! 🎓🥳
@zhaofengw
@sumanthd17
@BrihiJ
@kayo_yin
Stay tuned for news very soon… 👀
4
8
85
0
0
14
Schrödinger Project: The excitement of starting some new but also the inertia of actually doing it!!
1
1
13
No, this isn't us!!
But do checkout our work "FBI: Finding Blind Spots in Evaluator LLMs with Interpretable Checklists"
3
0
14
Everytime I’m waiting at baggage claim, I’m almost always convinced that they lost my baggage. Until ofc I find it 🙃
1
0
14
Well just realised that only 100 days left in this year. Now’s a good time to do Season 2
Any recommendations y’all?
I've done itttttt!! 100 days of reading. Ofc could've done some things better, but it's a start. A lot of learnings, many mistakes as well. I read decently diverse papers which helped me to develop lateral thinking* (100% would recommend)
More to come in the (Happy) New Year!!!
3
1
45
3
0
13
24
Seems like an important number
Definitely doesn’t feel like one though
Same old, same old
To bigger challenges and even bigger rewards 🥂
4
0
13
Phew!!! That was a tough ride
Finally done with one leg of the ride
📍Cochin - Alleppey - Varkala - Trivandrum - Kanyakumari
Now that the free trial is over, my new year start today! Looking forward to all the things that 2023 has to offer.
Clouds played spoil sport but such is life.
1
0
15
0
0
12
After almost 5months, back to cmd+v on a regular basis!!
No more of ctrl+shift+v
1
0
12
Every once in a while, pickup the damn phone and call your grandparents and the long-lost friends. Nothing gives more happiness than the blessings you get from them and ofc the "gaali" you get from friends. It was that day for me today!!!!
1
0
12
I pitched an Idea (for the first time). They liked it. Now we build on it 🚀. I want to come back to this tweet exactly 100 days from now and see where we are!!!
But first, Biryani 👨🍳
3
0
12
OKay! I guess I don't have to wake up every 30 mins tonight and check my mail :)
See ya tomorrow
The
#ACL2023NLP
committee is hard at work to finalize the submission decisions, which will be tentatively out by the end of May 1st (Anywhere on earth)....19 hours to go... Fingers crossed!
#ACL2023NLP
#ACL2023Toronto
9
22
200
0
0
12
AI4Bhart is announcing its first internship program. If you are interested in AI, passionate about building tools with technologies like Django and Javascript, and are motivated to contribute to the country's rich language tradition, please apply at [1/n]
1
6
12
From “Sumanth, don’t touch the tables” to “The tables look good!”
I finally feel worthy 😂
1
0
11
Half-way there! Some learnings
1. Speed at which I read has 📈 significantly
2. Able to connect ideas across different papers
3. Able to come up with analysis and viz. for my work
4. I feel more confident in *trying to* write a RW section
5. ⏫Confidence in asking Qs
[50/100]
On the Role of Bidirectionality in Language Model Pre-Training
[50/100]
Thankfully bi-directional models are not out of the race yet. My only concern is that moving masked tokens to the end for prediction - does that have an effect?
1
0
0
2
2
11
After messing up half my BTech to do what I'm doing now, I'm quite happy about the journey and thankful to many many people. [Narcissist Me - I hit a personal milestone today, and have been waiting for a while for this to happen]. To more and more 🥂
1
0
11
Note to self. The most important part of life is health.
SHOULD ALWAYS BE FIRST PRIORITY
2
0
11
Let's talk about MT eval!!
We're excited to have Nikita Moghe (
@nikita_moghe
) next week for our first talk of 2023!!
🗣️Title:"Taking Stock of Segment Level Evaluation in Machine Translation"
➡️Abstract:
📅Apr 19 @ 14:00 UTC
Links via mailing list:
1
12
46
0
0
11
🚨 New preprint alert 🚨
There is a cost associated with what adapters are brining to the table. In this work, we quantify these "costs" for NLU models (not LLMs)!
Checkout the summary of the paper in the thread👇
People have been yapping on and on about how adapters are this awesome efficient solution and that's cool. But do you have a moment to hear about our lords and saviours full-fine tuning and multi-task training?
A thread on our paper:
#NLProc
1
11
65
1
1
10
We’ll (
@gowtham_ramesh1
) be presenting Samanantar (TACL). In-person talk (25/05, session 7) & Poster (25/05, session 6)
0
2
10
If you are working on multilingual models/datasets, how do you find people for annotations? Asking for a friend
#NLProc
2
0
10
Excited to be attending the
@aclmeeting
in person! If you're heading to Toronto, let's connect and dive into discussions on all things multilingual.
#ACL2023
0
0
10
@bhogleharsha
said RCB is that team that submits the assignment after the deadline. But
@RCBTweets
showed that sometimes fighting for that 1% chance is good enough. Never been more proud of this team.
Go Aaaaaaarrrrrcccccbbbbbb!!!
“If there’s a 1% chance, it’s good enough” - VK
0
1
10
We are glad to announce the release of IndicGLUE evaluation benchmark on HuggingFace Datasets
Indic-BERT - …
You can access the IndicNLPSuite at -
@huggingface
#NLP
#DeepLearning
#AI
0
6
10
Mutter Paneer - Butter Chicken - Prawns Fry - Rotis (ofc no gol rotis 😪)
First one this year [1/12]
1
0
10
After a long time I was hoping to write a "🚨 Preprint Alert" thread
Damn Ilya, you stole my thunder :/
0
0
10
Naamapadam
Paper -
Data -
TL;DR - We created the largest NER corpora for 11 Indian languages, we created a human labelled test set and create SOTA NER models.
2
0
9
I guess no more pre-prints starting today :)
Those who know, will understand
0
0
9