Natasha Frumkin @icountfromzero.bsky.social · 02/05/2026I'm deeply grateful to my advisor @dianamarculescu.bsky.social for her guidance throughout this work, and to my colleagues in EnyAC at UT Austin for their continued encouragement. 000
Natasha Frumkin @icountfromzero.bsky.social · 02/05/2026📝 TL;DR: Q-Sched is a new post-training quantization paradigm that tunes the diffusion scheduler instead of the weights ⚡ — matching full-precision quality with a 4× smaller model 🚀, all without touching the backbone. 100
Natasha Frumkin @icountfromzero.bsky.social · 02/05/2026Excited to share that our paper "Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Scheduling" has been accepted to ICML 2026! 🎉 📄 Paper: arxiv.org/abs/2509.01624 💻 Code: github.com/enyac-group/... #ICML2026 #DiffusionModels #Quantization #EfficientML 100
Natasha Frumkin @icountfromzero.bsky.social · 03/09/2025Our work, Q-Sched, is compatible with state-of-the-art few-step diffusion models, achieving high fidelity images while only modifying the diffusion model scheduler. Special thanks to my research advisor, @dianamarculescu.bsky.social 010
Natasha Frumkin @icountfromzero.bsky.social · 03/09/2025We have released our latest work on quantizing few-step diffusion models using training-free scheduler adaptation! Arxiv: arxiv.org/pdf/2509.01624 Github: github.com/enyac-group/...arxiv.org 100
Reposted by Natasha FrumkinHung-Yueh Chiang @hychiang.bsky.social · 05/04/2025We’re excited to pre-release our latest work: Quamba2 🔧 Supports W4A8 / W4A16 / W4AX / W8A8 for Mamba1 and Mamba2 🚀 Achieves 4× memory reduction and 3× generation speedup ⚡️ Enables 8B model inference on Orin Nano 8G at 13 tokens/sec 🔥 Outperforms W4A8KV4 Llama3-8B in both speed and quality 121
Natasha Frumkin @icountfromzero.bsky.social · 29/03/2025We’re researching how people perceive visual artifacts in AI images — and your input would really help. It takes about 3 minutes. Participate here: t.co/WXuPdg9HKvt.cohttps://tinyurl.com/mtenmu9m 000
