vllm-omni/docs/user_guide/diffusion/attention_backends/huggingface_hub.md

Hugging Face Hub Attention Backends

Hub backends load the same published kernels commonly used with Hugging Face Diffusers. They are useful when training and serving must use matching kernel implementations to reduce numerical drift and sampling divergence.

Backend	Kernel	Platform
FLASH_ATTN_HUB	kernels-community/flash-attn2	Compatible CUDA GPUs
FLASH_ATTN_3_HUB	kernels-community/flash-attn3	Hopper sm_90 or newer

Installation

pip install kernels==0.14.1

If kernels is unavailable, vLLM-Omni warns and falls back to the corresponding local FlashAttention implementation. On CUDA devices older than Hopper, FLASH_ATTN_3_HUB falls back to FLASH_ATTN_HUB.

Usage

export DIFFUSION_ATTENTION_BACKEND=FLASH_ATTN_HUB

# Equivalent CLI selection
vllm-omni serve <model> --diffusion-attention-backend FLASH_ATTN_HUB

For per-role kernel alignment, configure the Hub backend on only the relevant role; see the attention backend overview.