FlashRT Adaptive Norms

This package provides FlashRT adaptive normalization kernels for Hugging Face Kernel Hub.

It is intended for DiT, VLA, video, and world-model blocks that use RMSNorm plus per-row style scale/shift/gate parameters.

Kernels

• ada_rms_norm_style_bf16: RMSNorm + style scale/shift, with BF16 output and BF16 gate extraction.
• gate_residual_ada_norm_fp8_static_bf16: in-place gated residual update, AdaRMSNorm, static FP8 E4M3 output, and BF16 gate extraction.

Hardware

• CUDA 12.8+
• BF16-capable NVIDIA GPUs
• FP8-capable NVIDIA GPUs for the FP8 output API

Current local source validation is on RTX 5090. Broader hardware rows should be added after installed-artifact validation.

Upstream

The serving source of truth remains FlashRT:

https://github.com/LiangSu8899/FlashRT