Ashwin Sankar

@_iunravel

MS student @iitmadras || Researcher @AI4Bharat || 侘び寂び (wabi sabi)

Chennai, India

Joined August 2019

285Following

775Followers

Pinned

Ashwin Sankar@_iunravel · Jul 12

I'm excited to share that I'll be spending about a month in Europe this summer, hopping between Austria, Denmark, and the Netherlands. My trip is anchored by two fantastic conferences: @aclmeeting in Vienna and @ISCAInterspeech in Rotterdam. I'll also be present at @iwslt…

2.0K

Pinned

Ashwin Sankar@_iunravel · Jun 24

This is what I want to wake up to every day, but I'm sleeping on it. Pretty sure someone is going to implement this.

462

Ashwin Sankar@_iunravel · Jul 20

Monkeys in my campus are really smart. They stole my peanut butter (1kg). Loading jacked monkeys soon 😂.

227

Ashwin Sankar@_iunravel · Jul 20

If I was caught talking to my code, would others think that I'm speaking fluent parseltongue? If the others understand what I'm saying, then do we all belong in Slytherin?

133

Ashwin Sankar@_iunravel · Jul 19

I'm happy today because I got to know that someone is building on top of indic-parler-tts. The outputs also sound very natural. Rooting for them to succeed.

630

Ashwin Sankar Retweeted

David Hall@dlwh · Jul 17

The good folks at Google have been putting together some new tutorials here github.com/rcrowe-google/…

157

Ashwin Sankar@_iunravel · Jul 16

There you go folks: tice.news/tice-dispatch/… @SarvamAI will open-source the Sovereign LLM that it is building as a part of @OfficialINDIAai mission. Just waiting for it to drop sometime this year 😄. CC: @paraschopra @prajdabre @cneuralnetwork

_iunravel's tweet card. India’s open-source AI model ready to lead the next wave. Sarvam AI’s LLM under IndiaAI Mission sets the stage for local, inclusive tech growth. TICE Dispatch

2.0K

Ashwin Sankar@_iunravel · Jul 15

Looks like I will have to wait till my current subscription ends. Keep cooking @GoogleDeepMind.

AAshwin Sankar@_iunravel · Jul 15

I ❤️ @Google

155

Ashwin Sankar@_iunravel · Jul 15

I ❤️ @Google

GGoogle India@GoogleIndia · Jul 15

If you’re a student in India - you’ve just been granted access to a FREE Gemini upgrade worth ₹19,500 for one year 🥳✨ Claim and get free access to Veo 3, Gemini in Google apps, and 2TB storage 🔗 goo.gle/freepro. @GeminiApp

468

Ashwin Sankar@_iunravel · Jul 13

Peeps stop using IndicTrans2. Use the Rotary IndicTrans2 which supports longer context, and it just got better for inference. Or use Sarvam-translate (IndicTrans3) which can translate academic papers and documents extremely well.

VVarun Gumma@VarunGumma23 · Jul 13

[1/N] Thanks to @Adalat_AI , the IndicTrans2-RoPE models are now CT2 compatible. Big shoutout to their entire team for finding our work interesting and porting them, which definitely wasn't easy (huggingface.co/collections/ad…)

409

Ashwin Sankar Retweeted

neural nets.@cneuralnetwork · Jul 12

one thing I genuinely hate about the Indian open source AI/tech scene is the attitude like genuinely when you are releasing a product why should you trash the earlier products? the thing is that, all the stuff is open source. and India in its totallity has around 0.001% number…

193

12.0K

Ashwin Sankar@_iunravel · Jul 12

Congratulations on the big achievement. But, let's skip the salt. We’ve already built Indic-Parler-TTS — 24 languages, 8000+ hours, 80k+ downloads — open-source, commercially usable, and thriving. Efforts like @SarvamAI only amplify what the ecosystem needs: serious research…

DDheemanth Reddy@Dheemanthreddy_ · Jul 11

VEENA GOT MORE DOWNLOADS THAN EVERY FUNDED AI STARTUP IN INDIA COMBINED THANK YOU FOR YOUR ATTENTION TO THIS MATTER.

9.0K

Ashwin Sankar@_iunravel · Jul 3

Whyyyyy????

473

Ashwin Sankar@_iunravel · Jul 2

We’ve launched the official SLTC X account!

IIEEE Speech and Language Processing@ieeesltc · Jul 2

🚀 We’re excited to announce the launch of the official X account for the IEEE Speech and Language Technical Committee: @ieeesltc! Follow us for updates, discussions, and announcements in speech and language processing. Let’s connect and grow the community!

4.0K

Ashwin Sankar@_iunravel · Jul 1

Hmm... I wonder how I can use this with Gemma 3n. Ohh that's right you can make smaller models using mix n match and use it for (self) speculative decoding. CC: @cneuralnetwork one of the reasons you should be training a matformer for your foundational model.

RRaj Dabre@prajdabre · Jul 1

If you haven't already then do read the paper on speculative decoding. The purpose is simple: Sample from a smaller model (drafter) such that it falls in the same distribution as larger model (verifier). Why: It's cheaper and faster to sample K tokens from a smaller model and…

357

Ashwin Sankar Retweeted

Guilherme Penedo@gui_penedo · Jun 27

We have finally released the 📝paper for 🥂FineWeb2, our large multilingual pre-training dataset. Along with general (and exhaustive) multilingual work, we introduce a concept that can also improve English performance: deduplication-based upsampling, which we call rehydration.

427

237

74.0K

Ashwin Sankar@_iunravel · Jun 27

Man, I finally found someone who actually sees the sh** I send.

nneural nets.@cneuralnetwork · Jun 27

😁

154

13.0K

Ashwin Sankar@_iunravel · Jun 16

Huge thanks to @Yuchenj_UW and @hyperbolic_labs for the incredibly generous $100 GPU grant! This is a massive boost for our work. 🚀 Seriously impressed by Hyperbolic Labs' commitment to accessible AI. They're currently offering the cheapest prices on the market for H100s;…

743