ipex-llm/python/llm/example/GPU/HuggingFace/Multimodal/StableDiffusion/README.md

Stable Diffusion

In this directory, you will find examples on how to run StableDiffusion models on Intel GPUs.

1. Installation

1.1 Install IPEX-LLM

Follow the instructions in IPEX-GPU installation guides (Linux Guide, Windows Guide) according to your system to install IPEX-LLM. After the installation, you should have created a conda environment, named diffusion for instance.

1.2 Install dependencies for Stable Diffusion

Assume you have created a conda environment named diffusion with ipex-llm installed. Run below commands to install dependencies for running Stable Diffusion.

conda activate diffusion
pip install diffusers["torch"] transformers
pip install -U PEFT transformers

2. Configures OneAPI environment variables for Linux

Note

Skip this step if you are running on Windows.

This is a required step on Linux for APT or offline installed oneAPI. Skip this step for PIP-installed oneAPI.

source /opt/intel/oneapi/setvars.sh

3. Runtime Configurations

For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device.

3.1 Configurations for Linux

For Intel Arc™ A-Series Graphics and Intel Data Center GPU Flex Series

export USE_XETLA=OFF
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
export SYCL_CACHE_PERSISTENT=1

For Intel Data Center GPU Max Series

export LD_PRELOAD=${LD_PRELOAD}:${CONDA_PREFIX}/lib/libtcmalloc.so
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
export SYCL_CACHE_PERSISTENT=1
export ENABLE_SDP_FUSION=1

For Intel iGPU

export SYCL_CACHE_PERSISTENT=1
export BIGDL_LLM_XMX_DISABLED=1

3.2 Configurations for Windows

For Intel iGPU

set SYCL_CACHE_PERSISTENT=1
set BIGDL_LLM_XMX_DISABLED=1

For Intel Arc™ A-Series Graphics

set SYCL_CACHE_PERSISTENT=1

Note

For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.

4. Examples

4.1 Openjourney Example

The example shows how to run Openjourney example on Intel GPU.

python ./openjourney.py

Arguments info:

• --repo-id-or-model-path REPO_ID_OR_MODEL_PATH: argument defining the huggingface repo id for the Openjourney model (e.g. prompthero/openjourney) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be 'prompthero/openjourney'.
• --prompt PROMPT: argument defining the prompt to be infered. It is default to be 'An astronaut in the forest, detailed, 8k'.
• --save-path: argument defining the path to save the generated figure. It is default to be openjourney-gpu.png.
• --num-steps: argument defining the number of inference steps. It is default to be 20.

4.2 StableDiffusion XL Example

The example shows how to run StableDiffusion XL example on Intel GPU.

python ./sdxl.py

Arguments info:

• --repo-id-or-model-path REPO_ID_OR_MODEL_PATH: argument defining the huggingface repo id for the stable diffusion xl model (e.g. stabilityai/stable-diffusion-xl-base-1.0) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be 'stabilityai/stable-diffusion-xl-base-1.0'.
• --prompt PROMPT: argument defining the prompt to be infered. It is default to be 'An astronaut in the forest, detailed, 8k'.
• --save-path: argument defining the path to save the generated figure. It is default to be sdxl-gpu.png.
• --num-steps: argument defining the number of inference steps. It is default to be 20.

The sample output image looks like below.

4.3 LCM-LoRA Example

The example shows how to performing inference with LCM-LoRA on Intel GPU.

python ./lora-lcm.py

Arguments info:

• --repo-id-or-model-path REPO_ID_OR_MODEL_PATH: argument defining the huggingface repo id for the stable diffusion xl model (e.g. stabilityai/stable-diffusion-xl-base-1.0) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be 'stabilityai/stable-diffusion-xl-base-1.0'.
• --lora-weights-path: argument defining the huggingface repo id for the LCM-LoRA model (e.g. latent-consistency/lcm-lora-sdxl) to be downloaded, or the path to huggingface checkpoint folder. It is default to be 'latent-consistency/lcm-lora-sdxl'.
• --prompt PROMPT: argument defining the prompt to be infered. It is default to be 'A lovely dog on the table, detailed, 8k'.
• --save-path: argument defining the path to save the generated figure. It is default to be lcm-lora-sdxl-gpu.png.
• --num-steps: argument defining the number of inference steps. It is default to be 4.