[DO NOT MERGE]Add tutorial for stable diffusion

[Stonepia]

This is a PR for Stable Diffusion tutorial.
Will put into the Dev Discuss, and it is for review only.

For the GitHub auto-rendered doc, please see https://github.com/intel/torch-xpu-ops/blob/tongsu/sd_tutorial/docs/tutorials/stable_diffusion.ipynb

Co-author: @gaopengff

[Stonepia]

BKC for reproducing the perf on PyTorch 2.6 vs PyTorch 2.7:

Copy and create the stable_diffusion.py from the last section of https://github.com/intel/torch-xpu-ops/blob/tongsu/sd_tutorial/docs/tutorials/stable_diffusion.ipynb
You could also use the one in the attachment and rename it to .py.

PT2.7:

conda create -n env_2.7 python=3.10
conda activate env_2.7
python -m pip install --pre torch==2.7.0.dev20250312+xpu --index-url https://download.pytorch.org/whl/nightly/xpu
pip install datasets accelerate einops torchmetrics transformers diffusers

# SD-v1-5 LCM
python -u stable_diffusion.py -m SimianLuo/LCM_Dreamshaper_v7 --precision fp16 --num_inference_steps 4 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3

# SDXL
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 512 --width 512 --warmup_iter 1 --iteration 3

# Additional Tests 
python -u stable_diffusion.py -m runwayml/stable-diffusion-v1-5 --precision fp16 --num_inference_steps 20 --height 512 --width 512 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m CompVis/stable-diffusion-v1-4 --precision fp16 --num_inference_steps 20 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 768 --width 768 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m stabilityai/stable-diffusion-2-1 --precision fp16 --num_inference_steps 20 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3

# Run img2img mode
python -u stable_diffusion.py -m SimianLuo/LCM_Dreamshaper_v7 --precision fp16 --num_inference_steps 4 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3 --pipeline_mode img2img
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 512 --width 512 --warmup_iter 1 --iteration 3 img2img

PT 2.6:

conda create -n env_2.6 python=3.10
conda activate env_2.6
pip install --pre torch==2.6.0 --index-url https://download.pytorch.org/whl/xpu
pip install datasets accelerate einops torchmetrics transformers diffusers

# SD-v1-5 LCM
python -u stable_diffusion.py -m SimianLuo/LCM_Dreamshaper_v7 --precision fp16 --num_inference_steps 4 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3
# SDXL
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 512 --width 512 --warmup_iter 1 --iteration 3

# Additional Tests 
python -u stable_diffusion.py -m runwayml/stable-diffusion-v1-5 --precision fp16 --num_inference_steps 20 --height 512 --width 512 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m CompVis/stable-diffusion-v1-4 --precision fp16 --num_inference_steps 20 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 768 --width 768 --warmup_iter 1 --iteration 3
python -u stable_diffusion.py -m stabilityai/stable-diffusion-2-1 --precision fp16 --num_inference_steps 20 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3

# Run img2img mode
python -u stable_diffusion.py -m SimianLuo/LCM_Dreamshaper_v7 --precision fp16 --num_inference_steps 4 --gscale 8.0 --height 512 --width 512 --warmup_iter 1 --iteration 3 --pipeline_mode img2img
python -u stable_diffusion.py -m stabilityai/stable-diffusion-xl-base-1.0 --precision fp16 --num_inference_steps 20 --gscale 7.5 --height 512 --width 512 --warmup_iter 1 --iteration 3 img2img

You will find that the output.csv and all the results are in it.
stable_diffusion.txt

[Copilot]

Copilot reviewed 1 out of 1 changed files in this pull request and generated no comments.