Our 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
03.09.2025 19:58 β π 1 π 0 π¬ 0 π 1
We 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/...
03.09.2025 19:57 β π 0 π 0 π¬ 1 π 0
Weβre excited to pre-release our latest work: Quamba2
π§ Supports W4A8 / W4A16 / W4AX / W8A8 for Mamba1 and Mamba2
π Achieves 4Γ memory reduction and 3Γ generation speedup
β‘οΈ Enables 8B model inference on Orin Nano 8G at 13 tokens/sec
π₯ Outperforms W4A8KV4 Llama3-8B in both speed and quality
05.04.2025 17:27 β π 3 π 1 π¬ 1 π 1
https://tinyurl.com/mtenmu9m
Weβre researching how people perceive visual artifacts in AI images β and your input would really help.
It takes about 3 minutes.
Participate here: t.co/WXuPdg9HKv
29.03.2025 00:29 β π 1 π 0 π¬ 0 π 0