Skip to content

README.md

Latest commit

 

History

History
50 lines (39 loc) · 2.86 KB

README.md

File metadata and controls

50 lines (39 loc) · 2.86 KB

xDiT Diffusion inference

rocm/pytorch-xdit images support several diffusion model inference workloads on gfx942 and gfx950 series (AMD Instinct™ MI300X, MI308X, MI325X and MI350X, MI355X) GPUs. The image has ROCm preview (based on TheRock) and uses xDiT distributed diffusion model inference framework for high-performance image and video generation.

Setup

Use

git clone https://github.com/ROCm/MAD
cd MAD
pip install -r requirements.txt

to clone the ROCm Model Automation and Dashboarding (MAD) repository to a local directory and install the required packages on the host machine.

Run MAD benchmarks

Execute benchmarks with

MAD_SECRETS_HFTOKEN=HFTOKEN madengine run --tags TAG --live-output

where HFTOKEN is a valid Hugging Face token (note that some models are gated) and TAG a supported MAD model tag. See Available models section for more details. The inference latencies can be found from results.csv once the benchmark runs have finished.

Available models

MAD model TAG Model repository
pyt_xdit_flux Flux.1
pyt_xdit_flux_kontext Flux.1 Kontext
pyt_xdit_flux_2 Flux.2
pyt_xdit_flux_2_klein Flux.2 Klein
pyt_xdit_hunyuanvideo HunyuanVideo
pyt_xdit_hunyuanvideo_1_5 HunyuanVideo 1.5
pyt_xdit_ltx_2 LTX-2
pyt_xdit_qwen_image Qwen-Image
pyt_xdit_qwen_image_edit Qwen-Image-Edit
pyt_xdit_sd_3_5 Stable diffusion 3.5
pyt_xdit_wan_2_1 Wan 2.1
pyt_xdit_wan_2_2 Wan 2.2
pyt_xdit_z_image Z-Image