PhD @LTIatCMU | Prev. @GoogleResearch @MSFTResearch
- Tweets 632
- Following 1,490
- Followers 4,113
- Likes 1,903
Pinned Tweet
I'm on the industry job market this year!
I'm a PhD student at CMU LTI. My work spans post-training, evals, alignment and multimodality!
Please reach out (DM) if you are hiring and think I'd be a good fit, I'd love to chat :)
Website: simran-khanuja.github.io/
CV: simran-khanuja.github.io/ass…
11
69
484
41,103
Simran Khanuja retweeted
I'm honored to be selected as a Rising Star in Data Science! datascience.uchicago.edu/res…
Excited to meet + learn from this great cohort in Chicago next week :)
15
6
99
3,491
Simran Khanuja retweeted
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over.
And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams.
So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super happy to finally release "The Ultimate MoE Training Handbook", free and open, organized around the questions training teams hit:
- how many experts, how big, shared or not?
- why is half my step spent in all-to-all, and how do I hide it?
- why is my run 25x slower with no error in the logs?
- softmax, sigmoid or sqrt(softplus) router, and why keep it in fp32?
- aux loss or DeepSeek's loss-free bias?
- upcycle a dense checkpoint or train from scratch?
- can I serve with fewer experts per token than I trained with?
⚠️ Warning: it's long (~70k words, budget a few days, sorry) but with 55 interactive figures, code you can run and numbers from our own runs.
Read it here: huggingface.co/spaces/huggin…
132
604
4,470
254,155
I'm on the industry job market this year!
I'm a PhD student at CMU LTI. My work spans post-training, evals, alignment and multimodality!
Please reach out (DM) if you are hiring and think I'd be a good fit, I'd love to chat :)
Website: simran-khanuja.github.io/
CV: simran-khanuja.github.io/ass…
11
69
484
41,103
For a more candid take on what I want to work on, check out my blog! simran-khanuja.github.io/blo…
1
7
1,862
Check out Rohit's <1 minute video which perfectly summarizes our recent work!!
Image generation models are hard to explore. Not anymore!
We present T-DIG, an LLM that learns ANY text-to-image model. Yes, ANY! GAN, diffusion, open, closed source!
You do NOT need access to weights of image model.
Ask for “varied backgrounds” and control how much variation.
3
8
1,339
I’m on the job market, and I keep getting asked: “What do you want to work on?”
I took some time to reflect on the art of problem selection in AI research and how that's evolved for me personally since 2017 :')
I'm no expert but if any of this resonates, lmk!
simran-khanuja.github.io/blo…
2
12
135
5,805
Asking an image model for Halloween costume ideas and getting the same witch 4-8 times?
Checkout T-DIG: a tool to explore variations inside any model for any request like "low-effort ones for ML researchers"!
Demo: image-study.org/demo
Gallery: image-study.org/gallery
(1/n)
3
12
71
10,362
I'd had this idea for a while and discussed it with @rohitgandikota earlier this year but we couldn't make progress until a good eval metric like TPIPS came out :’) Thanks to @ShengYuWang6 for a very timely release!
It was also great working with @GauravTParmar whose prior experience working on this problem was really helpful to learn from!
And lastly, thank you @gneubig for your support throughout!
And thanks to everyone who tried the tool in our study!
(10/n)
1
2
290
Website: image-study.org/t-dig/
Paper: arxiv.org/abs/2610.29003
Policies & data: huggingface.co/collections/n…
Code: github.com/simran-khanuja/t-…
Demo: image-study.org/demo/
Gallery: image-study.org/gallery
(fin)
2
270
yayyyy congrats!!
So honored and grateful to receive the Best Resource Paper award 🏅 for this work at #EMNLP26!! Thank you @emnlpmeeting for the recognition 🥰
Huge shoutout to all my amazing collaborators: Jena D. Hwang, @elisazmq_zheng, Jenna Godsey, @hyunw_kim,
@valentina__py, @_jennhu, and especially to my most amazing and supportive advisor @MaartenSap 🎊
12
1,258
