@RisingSayak
We've been studying what it takes to get NVFP4 & MXFP8 deliver good speedups on modern flow models for image & video gen. on B200 🕵️♂️ Today, I'm excited to share those findings! Bringing some cool recipes through Diffusers and TorchAO with `torch.compile` 🔥 Hop in ⬇️ https://t.co/gSd1Kwnu0l