Pocket powerhouse admist I/O awesomeness! Gemma 3n E4B & E2B are insane models, optimized for on-device while rivaling frontier models. It's a 🪆Matryoshka Transformer (MatFormer)🪆: Natively elastic b/w 4B & 2B pareto-optimally! ⭐️: free models with ZERO training cost! 🧵👇

Matryoshka-powered elasticity for the win! 🪆🚀 Excited to try Gemma 3n on my next smart fridge 🤓

Started from: seeing saturating retrieval accuracy curves with increasing embedding size on ImageNet after training a *single* ResNet-50 Now we here: "MRL Support" as a column in state-of-the-art embedding models 🥹🪆

Stop by CVPR poster #330 in Hall D for a chat about heterogenous and culturally representative generative models!

Check out this cool work from our lab!

did a (big) thing 🙃 still going to be running my brand practice, but now outta london! looking to take on projects in jun/jul (+ beyond), so if there's founders, investors, tech ppl u know and would like to connect me with, that'd be v kind <3 work at snehasanks.com

Had a fantastic time at #CVPR2025 and my spotlight talk on culturally representative T2I models at the DemoDiv workshop was definitely the highlight Thanks Polina Kirichenko and all the organizers, attendees and panelists for a super engaging & thought provoking workshop!

LLaVA-Prumerge, the first work of Visual Token Reduction for MLLM, finally got accepted after being cited 146 times since last year. Congrats to the team! Yuzhang Shang Yong Jae Lee See how to do MLLM inference much cheaper while holding performance. llava-prumerge.github.io

Culturally representative text to image models are surfing in to #iccv2025 🌊🏖️☀️ see you all in Hawaii!!