Shradha Sehgal

ML Research @Netflix | Prev @Google, @CarnegieMellon | Grad: UIUC, IIIT-Hyderabad | @SiebelScholars’24

[Shradha Sehgal @shradhasgl]

@rronak_ @MichaelElabd @QuantumArjun Congrats Ronak!

[Ahan Gupta @AhanGupta13]

1/🚨 A bit late to the party, but excited to share our paper at #ICLR2026:
AutoSP — the first compiler-based solution to unlock long-context LLM training.
✅ Up to 2.7× longer contexts on NVIDIA (2.5x on AMD)
✅ Negligible runtime overhead
✅ Merged into DeepSpeed 🚀
🧵👇

[Shradha Sehgal @shradhasgl]

I’ll be at NeurIPS from Dec 1-8! 🌊🏖️
Excited to catch up with old and new friends! Let’s grab coffee ☕️ or have a quick chat if you’re attending! #NeurIPS2025

[Shradha Sehgal @shradhasgl]

@krandiash Sonic! This is awesome! Congrats Karan and team.

[Nirav Diwan @ocean_drifters]

Excited to finally share our #NeurIPS2025 paper "🔮PurpCode: Reasoning for Safer Code Generation"!  🙌

👐 First post-training recipe for training safe code reasoning models
🚀 SOTA for cybersafety + utility, outperforming Sonnet 4, o4-mini, R1
🥇 Winner of 2025 Amazon Nova AI Challenge

📝 Paper: https://t.co/9aOoiL5zgJ

🧵👇 1/11

[Pratik Sampat @pratikrsampat]

Excited to finally share our new paper titled "CPU Autoscaling With a Kernel of Truth" at APSys 2025.

Huge thanks to my amazing collaborators and advisors @tianyin_xu, @SaugataGhose

Catch my virtual talk tonight if you're in Seoul!

Paper: https://t.co/xqwwu6NWiF

[Shradha Sehgal @shradhasgl]

It’s wild how much chatGPT has supercharged learning… I use it regularly to understand topics across fields like biology, economics, politics, etc, which ordinarily would take me hours of research & browsing

[Ishika Agarwal @wonderingishika]

I'm excited to announce NN-CIFT got into @NeurIPSConf 2025 (featuring a fancy, new title)💃💃🌴Can't wait to discuss it with everyone!!

Thank you @dilekhakkanitur and @convai_uiuc 🎉🎉

[Shradha Sehgal @shradhasgl]

@OfficialTanvi Congrats!!!! 🎊🍾🎈

[Revanth Gangi Reddy @gangi_official]

I've successfully defended my PhD thesis on automated information seeking! Extremely grateful to my advisor @hengjinlp, committee members and all collaborators.

Next, I'll be joining @GoogleDeepMind as a research scientist!

Link to defense slides: https://t.co/O4y30tnncC

[Shradha Sehgal @shradhasgl]

@gangi_official @hengjinlp @GoogleDeepMind Congrats! 🎉

[Shradha Sehgal @shradhasgl]

Amazing work! 😯

[Minjia Zhang @_Minjia_Zhang_]

🔬Interested in training AlphaFold3 faster, at scale, and beyond NVIDIA GPU? Now you can.

AlphaFold3 is a major leap in biomolecular modeling, but behind the scenes, it introduces severe system bottlenecks:

🧠 2D EvoAttention spikes memory usage

📉 Retrieval-augmented training pipeline causes long GPU idle time

⛔ Frequent but memory-intensive ops slow everything down

Today, I'm excited to announce MegaFold, a fully open-source system to make AlphaFold3 training fast, scalable, and cross-platform on both NVIDIA and AMD GPUs.

MegaFold delivers:

⚡ Up to 1.73x / 1.62x faster training on NVIDIA H100 / AMD MI250

🧬 Up to 1.35× longer sequences compared to PyTorch baseline

Key features:

🚀 Memory-Efficient EvoAttention via portable Triton kernels

💡 Ahead-of-Time Caching to eliminate GPU idle time in retrieval pipelines

🔗 DeepFusion for reducing overhead of small but frequent memory-intensive AF3 ops

📘 Project page: https://t.co/sSaxDtBT1O

📄 Paper: https://t.co/2sCArFoq82

💻 Code: https://t.co/gn7HR5kcQC

🤝 MegaFold is developed in collaboration between UIUC SSAIL Lab and researchers from University of Missouri and Lawrence Berkeley National Laboratory.

Kudos to the brilliant team: Hoa La, Ahan Gupta, Alex Morehead, Jianlin Cheng

#AlphaFold3 #AI #ProteinFolding #Bioinformatics #AMD #Triton #CrossPlatform #OpenSource

[Minjia Zhang @_Minjia_Zhang_]

🔬Interested in training AlphaFold3 faster, at scale, and beyond NVIDIA GPU? Now you can.

AlphaFold3 is a major leap in biomolecular modeling, but behind the scenes, it introduces severe system bottlenecks:

🧠 2D EvoAttention spikes memory usage

📉 Retrieval-augmented training pipeline causes long GPU idle time

⛔ Frequent but memory-intensive ops slow everything down

Today, I'm excited to announce MegaFold, a fully open-source system to make AlphaFold3 training fast, scalable, and cross-platform on both NVIDIA and AMD GPUs.

MegaFold delivers:

⚡ Up to 1.73x / 1.62x faster training on NVIDIA H100 / AMD MI250

🧬 Up to 1.35× longer sequences compared to PyTorch baseline

Key features:

🚀 Memory-Efficient EvoAttention via portable Triton kernels

💡 Ahead-of-Time Caching to eliminate GPU idle time in retrieval pipelines

🔗 DeepFusion for reducing overhead of small but frequent memory-intensive AF3 ops

📘 Project page: https://t.co/sSaxDtBT1O

📄 Paper: https://t.co/2sCArFoq82

💻 Code: https://t.co/gn7HR5kcQC

🤝 MegaFold is developed in collaboration between UIUC SSAIL Lab and researchers from University of Missouri and Lawrence Berkeley National Laboratory.

Kudos to the brilliant team: Hoa La, Ahan Gupta, Alex Morehead, Jianlin Cheng

#AlphaFold3 #AI #ProteinFolding #Bioinformatics #AMD #Triton #CrossPlatform #OpenSource

[Shradha Sehgal @shradhasgl]

@AkshayGoindani1 Thanks @AkshayGoindani1 looks like we might be quick to assume these gains since evils did not reproduce the baselines effectively

https://t.co/0LEygcoxJl

[Shashwat Goel @ ICML 🇰🇷 @ShashwatGoel7]

Confused about recent LLM RL results where models improve without any ground-truth signal? We were too. Until we looked at the reported numbers of the Pre-RL models and realized they were serverely underreported across papers. We compiled discrepancies in a blog below🧵👇

[Shradha Sehgal @shradhasgl]

Can someone pls give a tldr of what happened in RLVR this past week…

[Shradha Sehgal @shradhasgl]

Such interesting insights!! 😮 Esp for cultural linguistic analysis of LLMs

[Ishika Agarwal @wonderingishika]

Would models know more about Indian food in Hindi and Turkey’s history in Turkish? Does the language of a question affect an LLM’s answer?

✨Yes!✨

@nbbozdag and I are excited to announce our newest preprint in which we explore “Language Specific Knowledge (LSK)”.

[Shradha Sehgal @shradhasgl]

@lmathur_ @GoogleDeepMind Congrats Leena!

[Shradha Sehgal @shradhasgl]

no I am not moving yet, just polling 🙂‍↔️

[Shradha Sehgal @shradhasgl]

For ppl who moved from South Bay to SF or vice versa, what made u do it? What did you like / dislike? and would u recommend it?

[Shradha Sehgal @shradhasgl]

@adityakusupati @_keysarasara @X @UW Congratulations!!! 🎊