### Setup pre-commit hooks

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install pre-commit and set up hooks for code formatting and linting.

```bash
pip3 install pre-commit
pre-commit install
pre-commit run --all-files
```

--------------------------------

### Start ComfyUI

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/OnlineQuantization.md

Command to start ComfyUI, optimized for GPU usage.

```bash
python main.py --gpu-only
```

--------------------------------

### Install WSL2

Source: https://github.com/siliconflow/onediff/wiki/Run-OneDiff-on-Windows-by-WSL2

Commands to install and configure the Windows Subsystem for Linux.

```bash
wsl --install
```

```bash
dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart
```

```bash
dism.exe /online /enable-feature /featurename:VirtualMachinePlatform /all /norestart
```

```bash
wsl --set-default-version 2
```

--------------------------------

### Install onediff from source

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Navigate to the cloned onediff directory and install it using pip.

```bash
cd onediff && python3 -m pip install -e .
```

--------------------------------

### Install onediff for development

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install onediff from source with development dependencies.

```bash
cd onediff && python3 -m pip install -e '.[dev]'
```

--------------------------------

### Run Examples

Source: https://github.com/siliconflow/onediff/blob/main/CONTRIBUTING.md

Execute example scripts to verify functionality before proceeding with release steps. Ensure you are in the correct directory.

```bash
cd onediff_diffusers_extensions
python3 examples/text_to_image.py
```

--------------------------------

### Install Build Tools

Source: https://github.com/siliconflow/onediff/blob/main/CONTRIBUTING.md

Install the necessary Python packages for building and uploading distribution packages.

```bash
python3 -m pip install build twine
```

--------------------------------

### Install Dependencies and Build Docker Image

Source: https://github.com/siliconflow/onediff/blob/main/benchmarks/README.md

Installs project dependencies and builds the OneDiff Docker image. Ensure you have Python 3 and pip installed.

```bash
python3 -m pip install -r requirements.txt
python3 docker/main.py --yaml ./docker/config/community-default.yaml
```

--------------------------------

### Install PyTorch and Diffusers

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install the required versions of PyTorch, transformers, and diffusers libraries.

```bash
python3 -m pip install "torch" "transformers==4.27.1" "diffusers[torch]==0.19.3"
```

--------------------------------

### Start sd-webui Client with API

Source: https://github.com/siliconflow/onediff/blob/main/onediff_sd_webui_extensions/README.md

Run this command to start the Stable Diffusion WebUI client with API enabled, allowing external requests.

```bash
# start sd-webui client
python3 webui.py --api
```

--------------------------------

### Install OneDiffX

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/README.md

Clone the OneDiff repository and install the onediffx package from the cloned directory.

```bash
git clone https://github.com/siliconflow/onediff.git
cd onediff_diffusers_extensions && python3 -m pip install -e .
```

--------------------------------

### OneDiff version format example

Source: https://github.com/siliconflow/onediff/wiki/How-to-Run-or-Debug-OneDiff

Example output showing the standard versioning format for OneDiff and OneDiffX.

```text
OneDiff version: 1.1.1.dev63+gb1246aaf.d20240726
OneDiffX version: 1.1.1.dev63+gb1246aaf.d20240726
```

--------------------------------

### Install ComfyUI and OneDiff

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/sd3/README.md

Installs ComfyUI and the onediff & onediff_comfy_nodes packages. Ensure Python 3.10 is used. This setup links the onediff custom nodes into the ComfyUI directory.

```shell
# python 3.10
COMFYUI_DIR=$pwd/ComfyUI
# install ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git

# install onediff & onediff_comfy_nodes
git clone https://github.com/siliconflow/onediff.git
cd onediff && pip install -r onediff_comfy_nodes/sd3/requirements.txt && pip install -e .
ln -s $pwd/onediff/onediff_comfy_nodes  $COMFYUI_DIR/custom_nodes
```

--------------------------------

### Install Comfy CLI

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/README.md

Installs the Comfy CLI tool. This is the first step for installing ComfyUI via the CLI method.

```shell
pip install comfy-cli
```

--------------------------------

### Install cubiq/ComfyUI_InstantID

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/README.md

Clones the repository and resets to a specific commit hash for version stability.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/cubiq/ComfyUI_InstantID.git
# Date:   Thu Jul 11 19:26:22 2024 +0200
git reset --hard  6d95aa6758e58dab550725e59dcefbd426c160c7
```

--------------------------------

### Install Nexfort Compiler Backend

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install the optional Nexfort compiler backend and compatible PyTorch versions.

```bash
python3 -m  pip install -U torch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 torchao==0.1
python3 -m  pip install -U nexfort
```

--------------------------------

### Install Nexfort Backend

Source: https://github.com/siliconflow/onediff/blob/main/src/onediff/infer_compiler/backends/nexfort/README.md

Installs the nexfort package. Requires a compatible PyTorch and CUDA environment.

```bash
# The current version of nexfort is compatible with torch 2.3.0.
pip3 install nexfort
```

--------------------------------

### Install cubiq/ComfyUI_IPAdapter_plus

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/README.md

Clones the repository and resets to a specific commit hash for version stability.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/cubiq/ComfyUI_IPAdapter_plus.git
# Date:   Thu Jul 18 15:12:49 2024 +0200
git reset --hard  78ac59c61c8caf33e3419d2c8f70838b2da0fb04
```

--------------------------------

### Install onediff from PyPI

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install the onediff package from the Python Package Index (PyPI).

```bash
python3 -m pip install --pre onediff
```

--------------------------------

### Install cubiq/PuLID_ComfyUI

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/README.md

Clones the repository and resets to a specific commit hash for version stability.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/cubiq/PuLID_ComfyUI.git
# Date:   Sun May 26 19:23:51 2024 +0200
git reset --hard  80d157ace9e4d2db8c963e770ed99d492d0f4569
```

--------------------------------

### Run Stable Diffusion WebUI Service

Source: https://github.com/siliconflow/onediff/blob/main/onediff_sd_webui_extensions/README.md

Navigate to the stable-diffusion-webui directory and run the webui.sh script with the desired port to start the service.

```bash
cd stable-diffusion-webui
python webui.py --port 7860
```

--------------------------------

### Install PyTorch and Dependencies

Source: https://github.com/siliconflow/onediff/blob/main/src/onediff/infer_compiler/backends/nexfort/README.md

Installs PyTorch version 2.3.0 and related libraries. Ensure your environment is compatible before running.

```bash
pip3 install -U torch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 torchao==0.1
```

--------------------------------

### Install OneDiff and ComfyUI Nodes

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/benchmarks/README.md

Commands to clone the repository, install dependencies, and link the custom nodes to the ComfyUI directory.

```shell
# python 3.10
git clone https://github.com/siliconflow/onediff.git
cd onediff && pip install -e .
ln -s $(pwd)/onediff_comfy_nodes path/to/ComfyUI/custom_nodes/
# or
# cp -r onediff_comfy_nodes path/to/ComfyUI/custom_nodes/
```

--------------------------------

### Run Stable Diffusion WebUI

Source: https://github.com/siliconflow/onediff/wiki/Run-OneDiff-on-Windows-by-WSL2

Commands to navigate to the WebUI directory and start the application.

```bash
cd /path/to/your/stable-diffusion-webui
```

```bash
python webui.py
```

--------------------------------

### Install Stable Diffusion WebUI and OneDiff Extensions

Source: https://github.com/siliconflow/onediff/blob/main/onediff_sd_webui_extensions/README.md

Clone the repositories for Stable Diffusion WebUI and OneDiff, then link the OneDiff extension to the WebUI's extensions directory. Finally, install all of Stable Diffusion WebUI's dependencies.

```bash
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
git clone https://github.com/siliconflow/onediff.git
ln -s "$(pwd)/onediff/onediff_sd_webui_extensions" "$(pwd)/stable-diffusion-webui/extensions/onediff_sd_webui_extensions"
cd stable-diffusion-webui

# Install all of stable-diffusion-webui's dependencies.
# If you install as root user, append `-f` to the end of the command line.
venv_dir=- bash webui.sh
```

--------------------------------

### Initialize Graph Caching

Source: https://context7.com/siliconflow/onediff/llms.txt

Setup for saving compiled graphs to disk for production deployment.

```python
import torch
from diffusers import DiffusionPipeline
from onediff.infer_compiler import oneflow_compile, OneflowCompileOptions

pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16",
)
pipe.to("cuda")
```

--------------------------------

### Install ComfyUI-AnimateDiff-Evolved

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/README.md

Clones the repository and resets to a specific commit hash for version stability.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved.git
git reset --hard  1b660e535381a4b4e4028aad38b7ae21c81b861c
```

--------------------------------

### Run ComfyUI with OneDiff

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/benchmarks/README.md

Commands to install the benchmark environment and launch the ComfyUI server with the required extra model paths.

```shell
export COMFYUI_ROOT=path/to/ComfyUI

cd path/to/onediff/onediff_comfy_nodes/benchmarks

bash scripts/install_env.sh $COMFYUI_ROOT

cd  $COMFYUI_ROOT

python main.py --gpu-only --port 8188 --extra-model-paths-config path/to/onediff/tests/comfyui/extra_model_paths.yaml
```

--------------------------------

### Install Diffusers Library

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/flux/README.md

Ensure the diffusers library is installed with PyTorch support. This is a prerequisite for using diffusion models like Flux.

```bash
pip3 install --upgrade diffusers[torch]
```

--------------------------------

### Install Diffusers for Kolors

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/kolors/README.md

Install the specific version of the diffusers library required to support the Kolors pipeline.

```bash
# Ensure diffusers include the kolors pipeline.
pip install git+https://github.com/huggingface/diffusers.git
```

--------------------------------

### Install OneDiff Package

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/README.md

Installs the OneDiff package in editable mode after cloning the repository. This makes the OneDiff library available for use.

```shell
cd onediff && pip install -e .
```

--------------------------------

### Install OneDiff Enterprise (CN, CUDA 12.1)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in China with CUDA 12.1. Uses CN-specific repositories.

```bash
python3 -m pip install --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/main/cu121/ && \
python3 -m pip install --pre onediff-quant -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/onediff-quant/ && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Install OneDiff Enterprise (CN, CUDA 11.8)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in China with CUDA 11.8. Uses CN-specific repositories.

```bash
python3 -m pip install --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/main/cu118/ && \
python3 -m pip install --pre onediff-quant -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/onediff-quant/ && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Install ComfyUI_InstantID Custom Node

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/modules/oneflow/hijack_comfyui_instantid/README.md

Clone the ComfyUI_InstantID repository and reset to a specific commit. Ensure you are in the 'ComfyUI/custom_nodes' directory before running.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/cubiq/ComfyUI_InstantID.git
git reset --hard d8c70a0cd8ce0d4d62e78653674320c9c3084ec1
```

--------------------------------

### Install OneDiff Enterprise (CN, CUDA 12.2)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in China with CUDA 12.2. Uses CN-specific repositories.

```bash
python3 -m pip install --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/main/cu122/ && \
python3 -m pip install --pre onediff-quant -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/onediff-quant/ && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Run SDXL with DeepCache

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Requires OneDiffX installation. Executes the enterprise script with DeepCache enabled for SDXL models.

```bash
python text_to_image_deep_cache_sd_sdxl_enterprise.py --model $model_path --model_type sdxl --saved_image output_deepcache.png
```

--------------------------------

### Install OneDiff Extension for WebUI

Source: https://github.com/siliconflow/onediff/wiki/Run-OneDiff-on-Windows-by-WSL2

Commands to clone the OneDiff repository and link it to the Stable Diffusion WebUI extensions directory.

```bash
git clone https://github.com/siliconflow/onediff.git
```

```bash
ln -s "$(pwd)/onediff/onediff_sd_webui_extensions" "$(pwd)/stable-diffusion-webui/extensions/onediff_sd_webui_extensions"

cd $(pwd)/stable-diffusion-webui
venv_dir=- bash webui.sh 

# Exit webui server and upgrade some of the components that conflict with onediff.
cd repositories/generative-models && git checkout 9d759324 && cd -
```

--------------------------------

### Run SD with DeepCache

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Requires OneDiffX installation. Executes the enterprise script with DeepCache enabled for SD models.

```bash
python text_to_image_deep_cache_sd_sdxl_enterprise.py --model $model_path --model_type sd --saved_image output_deepcache_sd.png
```

--------------------------------

### Run Stable Diffusion XL with OneDiffX DeepCache

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/README.md

Initialize and compile a Stable Diffusion XL pipeline with DeepCache enabled for accelerated inference. This example sets up the pipeline and defines a prompt.

```python
import torch

from onediffx import compile_pipe
from onediffx.deep_cache import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
)
pipe.to("cuda")

pipe = compile_pipe(pipe)

prompt = "A photo of a cat. Focus light and create sharp, defined edges."

```

--------------------------------

### Install ComfyUI via CLI

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/README.md

Installs ComfyUI using the Comfy CLI. This command automates the ComfyUI installation process.

```shell
comfy install
```

--------------------------------

### Compile Arbitrary Models with OneDiff

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/sd3/README.md

Demonstrates compiling a custom PyTorch module using OneDiff's infer_compiler with the Nexfort backend. This example requires `is_nexfort_available` to be true and uses specific compilation options for optimization.

```python
# Compile arbitrary models (torch.nn.Module)
import torch
from onediff.utils.import_utils import is_nexfort_available
assert is_nexfort_available() == True

import onediff.infer_compiler as infer_compiler

class MyModule(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.lin = torch.nn.Linear(100, 10)

    def forward(self, x):
        return torch.nn.functional.relu(self.lin(x))

mod = MyModule().to("cuda").half()
with torch.inference_mode():
    compiled_mod = infer_compiler.compile(mod,
        backend="nexfort",
        options={"mode": "max-autotune:cudagraphs", "dynamic": True, "fullgraph": True},
    )
    print(compiled_mod(torch.randn(10, 100, device="cuda").half()).shape)

print("Successfully installed～")
```

--------------------------------

### Install OneDiff Enterprise (NA/EU, CUDA 12.2)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in North America/Europe with CUDA 12.2. Ensure you have the correct CUDA version installed.

```bash
python3 -m pip install --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/enterprise_cu122 && \
python3 -m pip install --pre onediff-quant -f https://github.com/siliconflow/onediff_releases/releases/expanded_assets/enterprise && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Set up Latte Environment

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/latte/README.md

Clone the Latte repository, navigate into the directory, and set the PYTHONPATH environment variable. This is necessary for using Latte models.

```bash
git clone -b run https://github.com/siliconflow/dit_latte/
cd dit_latte
export PYTHONPATH=`pwd`:$PYTHONPATH
```

--------------------------------

### Install OneDiff Enterprise (NA/EU, CUDA 12.1)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in North America/Europe with CUDA 12.1. Ensure you have the correct CUDA version installed.

```bash
python3 -m pip install --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/enterprise_cu121 && \
python3 -m pip install --pre onediff-quant -f https://github.com/siliconflow/onediff_releases/releases/expanded_assets/enterprise && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Install OneDiff Enterprise (NA/EU, CUDA 11.8)

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Installs OneDiff Enterprise and its dependencies for users in North America/Europe with CUDA 11.8. Ensure you have the correct CUDA version installed.

```bash
python3 -m pip install --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/enterprise_cu118 && \
python3 -m pip install --pre onediff-quant -f https://github.com/siliconflow/onediff_releases/releases/expanded_assets/enterprise && \
python3 -m pip install git+https://github.com/siliconflow/onediff.git@main#egg=onediff
```

--------------------------------

### Build Distribution Wheel

Source: https://github.com/siliconflow/onediff/blob/main/CONTRIBUTING.md

Clean the distribution directory and then build the wheel package for the project.

```bash
rm -rf dist
python3 -m build
```

--------------------------------

### Install Required Python Packages

Source: https://github.com/siliconflow/onediff/wiki/Run-OneDiff-on-Windows-by-WSL2

Install specific versions of dependencies required for OneDiff.

```bash
pip install -U einops==0.7.0
```

--------------------------------

### Download Models for Benchmarking

Source: https://github.com/siliconflow/onediff/blob/main/benchmarks/README.md

Downloads the necessary models for running the OneDiff benchmark. This script should be executed in the project's root directory.

```bash
sh download_models.sh models
```

--------------------------------

### Download SD 1.5 Quantization Info

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Use wget to download the quantization info file for SD 1.5 and place it in the ComfyUI onediff_quant directory.

```bash
wget https://huggingface.co/siliconflow/stable-diffusion-v1-5-onediff-comfy-enterprise-v1/resolve/main/v1-5-pruned_quantize_info.pt -O ComfyUI/models/onediff_quant/v1-5-pruned_quantize_info.pt
```

--------------------------------

### Install OneFlow for CUDA 12.1 (CN)

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install OneFlow for CUDA 12.1 using pip. This command is for users in China.

```bash
python3 -m pip install -U --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu122
```

--------------------------------

### Install OneFlow for CUDA 11.8 (CN)

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install OneFlow for CUDA 11.8 using pip. This command is for users in China.

```bash
python3 -m pip install -U --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu118
```

--------------------------------

### Download SD 2.1 Quantization Info

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Use wget to download the quantization info file for SD 2.1 and place it in the ComfyUI onediff_quant directory. Note: The target filename in the command is a placeholder and should match the actual file name after download and potential extraction.

```bash
wget https://huggingface.co/siliconflow/stable-diffusion-v2-1-onediff-comfy-enterprise/blob/main/v2-1_768-ema-pruned.pt -O ComfyUI/models/onediff_quant/v1-5-pruned_quantize_info.pt
```

--------------------------------

### Download SDXL Base Model and InstantID Models (NA/EU)

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/modules/oneflow/hijack_comfyui_instantid/README.md

Download the Stable Diffusion XL base model, InstantID adapter, and ControlNet model for NA/EU users. Place them in the specified directories within ComfyUI.

```shell
cd ComfyUI
# Load Checkpoint
wget -O models/checkpoints/sd_xl_base_1.0.safetensors https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

# Load InstantID Model
mkdir -p models/instantid
wget -O models/instantid/ip-adapter.bin https://huggingface.co/InstantX/InstantID/resolve/main/ip-adapter.bin


# Load ControlNet Model
wget -O models/controlnet/diffusion_pytorch_model.safetensors https://huggingface.co/InstantX/InstantID/resolve/main/ControlNetModel/diffusion_pytorch_model.safetensors
```

--------------------------------

### Install OneFlow for CUDA 12.1 (NA/EU)

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install OneFlow for CUDA 12.1 using pip. This command is for users in North America and Europe.

```bash
python3 -m pip install -U --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/community_cu122
```

--------------------------------

### Download SD 1.5 Checkpoint Model

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Use wget to download the pruned safetensors file for Stable Diffusion 1.5 and place it in the ComfyUI checkpoints directory.

```bash
wget https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned.safetensors -O ComfyUI/models/checkpoints/v1-5-pruned.safetensors
```

--------------------------------

### Install OneFlow for CUDA 11.8 (NA/EU)

Source: https://github.com/siliconflow/onediff/blob/main/README.md

Install OneFlow for CUDA 11.8 using pip. This command is for users in North America and Europe.

```bash
python3 -m pip install -U --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/community_cu118
```

--------------------------------

### Execute Text-to-Image Benchmark

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/benchmarks/README.md

Command to run the text-to-image benchmark script, with results saved to the results directory.

```shell
cd path/to/onediff/onediff_comfy_nodes/benchmarks

bash scripts/run_text_to_image.sh
```

--------------------------------

### Download SDXL Base Model and InstantID Models (CN)

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/modules/oneflow/hijack_comfyui_instantid/README.md

Download the Stable Diffusion XL base model, InstantID adapter, and ControlNet model for CN users using HF mirror. Place them in the specified directories within ComfyUI.

```shell
cd ComfyUI
wget -O models/checkpoints/sd_xl_base_1.0.safetensors https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

# Load InstantID Model
mkdir -p models/instantid
wget -O models/instantid/ip-adapter.bin https://hf-mirror.com/InstantX/InstantID/resolve/main/ip-adapter.bin

# Load ControlNet Model
wget -O models/controlnet/diffusion_pytorch_model.safetensors https://hf-mirror.com/InstantX/InstantID/resolve/main/ControlNetModel/diffusion_pytorch_model.safetensors
```

--------------------------------

### Run Quantization Benchmark

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/pixart/README.md

Use this command to run a benchmark with quantization enabled. It specifies the model, scheduler, output image, prompt, compiler, compiler configuration, and quantization configuration. Adjust parameters as needed for different models and quantization types.

```python
python3 ./benchmarks/text_to_image.py \
--model PixArt-alpha/PixArt-XL-2-1024-MS \
--scheduler none \
--steps 20 \
--output-image ./pixart_alpha_fp8.png \
--prompt "product photography, world of warcraft orc warrior, white background" \
--compiler nexfort \
--compiler-config '{"mode": "quant:max-optimize:max-autotune:freezing:benchmark:low-precision:cudagraphs", "memory_format": "channels_last"}' \
--quantize \
--quantize-config '{"quant_type": "fp8_e4m3_e4m3_dynamic"}'
```

--------------------------------

### Install OneDiff Comfy Nodes via CLI

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/README.md

Installs the OneDiff Comfy Nodes using the Comfy CLI. This command integrates OneDiff functionality into ComfyUI.

```shell
comfy node install onediff_comfy_nodes
```

--------------------------------

### Run SD1.5 Benchmark With NexFort Compilation

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd1.5/README.md

Execute the text-to-image benchmark using the NexFort compiler with specific optimization configurations.

```shell
python3 benchmarks/text_to_image.py \
  --model runwayml/stable-diffusion-v1-5 \
  --height 512 --width 512 \
  --scheduler none \
  --steps 20 \
  --output-image ./stable-diffusion-v1-5-compile.png \
  --prompt "beautiful scenery nature glass bottle landscape, , purple galaxy bottle," \
  --compiler nexfort \
  --compiler-config '{"mode": "cudagraphs:benchmark:max-autotune:low-precision:cache-all", "memory_format": "channels_last", "options": {"inductor.optimize_linear_epilogue": true, "overrides.conv_benchmark": true, "overrides.matmul_allow_tf32": true}}' \
  --seed 1 \
  --print-output
```

--------------------------------

### Check PyTorch Version

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/OnlineQuantization.md

Verify your PyTorch installation version.

```python
import torch; print(torch.__version__)
```

--------------------------------

### Download Model Files

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/modules/nexfort/hijack_comfyui_instantid/README.md

Scripts to download required checkpoints and models for InstantID, provided for different geographic regions.

```shell
cd ComfyUI
# Load Checkpoint
wget -O models/checkpoints/sd_xl_base_1.0.safetensors https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

# Load InstantID Model
mkdir -p models/instantid
wget -O models/instantid/ip-adapter.bin https://huggingface.co/InstantX/InstantID/resolve/main/ip-adapter.bin


# Load ControlNet Model
wget -O models/controlnet/diffusion_pytorch_model.safetensors https://huggingface.co/InstantX/InstantID/resolve/main/ControlNetModel/diffusion_pytorch_model.safetensors
```

```shell
cd ComfyUI
wget -O models/checkpoints/sd_xl_base_1.0.safetensors https://hf-mirror.com/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

# Load InstantID Model
mkdir -p models/instantid
wget -O models/instantid/ip-adapter.bin https://hf-mirror.com/InstantX/InstantID/resolve/main/ip-adapter.bin

# Load ControlNet Model
wget -O models/controlnet/diffusion_pytorch_model.safetensors https://hf-mirror.com/InstantX/InstantID/resolve/main/ControlNetModel/diffusion_pytorch_model.safetensors
```

--------------------------------

### Install Diffusers Library

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd35/README.md

Upgrade the diffusers library to ensure support for the SD35 pipeline.

```bash
# Ensure diffusers include the SD35 pipeline.
pip3 install --upgrade diffusers[torch]
```

--------------------------------

### Text-to-Image Generation with Quantization and T5 Speedup

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/flux/README.md

This command enables text-to-image generation with quantization and specifically speeds up the T5 model, which is crucial for memory-constrained GPUs like the RTX 4090.

```bash
NEXFORT_FORCE_QUANTE_ON_CUDA=1 python3 onediff_diffusers_extensions/examples/flux/text_to_image_flux.py \
    --quantize \
    --transform \
    --speedup-t5 \
    --saved-image flux_compile.png
```

--------------------------------

### Download SVD and SDXL Model Files

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/ComfyUI_Online_Quantization.md

Downloads the Stable Diffusion XL base and SVD XT model checkpoints into the ComfyUI models directory.

```bash
cd ComfyUI
wget -O  models/checkpoints/sd_xl_base_1.0.safetensors https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors
wget -O  models/checkpoints/svd_xt_1_1.safetensors https://huggingface.co/vdo/stable-video-diffusion-img2vid-xt-1-1/resolve/main/svd_xt_1_1.safetensors
```

--------------------------------

### Quantize Custom Models

Source: https://github.com/siliconflow/onediff/blob/main/src/onediff/quantization/README.md

Runs the test script to verify and perform quantization on custom models.

```bash
python tests/test_quantize_custom_model.py
```

--------------------------------

### Download SD 2.1 Checkpoint Model

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Use wget to download the SD 2.1 safetensors file (after unzipping the provided zip) and place it in the ComfyUI checkpoints directory.

```bash
wget https://huggingface.co/siliconflow/stable-diffusion-v2-1-onediff-comfy-enterprise/blob/main/v2-1_768-ema-pruned.zip -O ComfyUI/models/checkpoints/v2-1_768-ema-pruned.zip
```

--------------------------------

### Perform inference using native SDXL

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/docs/onediffx_lora_examples.ipynb

Initializes and compiles an SDXL pipeline for inference. Ensure the model path is correct and the environment is set up for CUDA.

```python
import torch
from diffusers import DiffusionPipeline
from onediffx import compile_pipe
from onediffx.lora import load_and_fuse_lora, set_and_fuse_adapters, delete_adapters, unfuse_lora, load_lora_and_optionally_fuse

MODEL_ID = "/share_nfs/hf_models/stable-diffusion-xl-base-1.0"
pipe = DiffusionPipeline.from_pretrained(MODEL_ID, variant="fp16", torch_dtype=torch.float16).to("cuda")
pipe = compile_pipe(pipe)

images_fusion = pipe(
    "a cat",
    height=1024,
    width=1024,
    generator=torch.manual_seed(0),
    num_inference_steps=30,
).images[0]
display(images_fusion.resize((512, 512)))
```

--------------------------------

### Accelerate SD3.5 Text-to-Image with Quantization and T5 Speedup

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd35/README.md

This command accelerates SD3.5 text-to-image generation, quantizing the T5 model for reduced memory usage on GPUs like the RTX 4090 with limited VRAM. The `--speedup-t5` flag is crucial for memory-constrained environments.

```bash
NEXFORT_FORCE_QUANTE_ON_CUDA=1 python3 onediff_diffusers_extensions/examples/sd35/text_to_image_sd35.py \
    --quantize \
    --transform \
    --speedup-t5 \
    --saved-image sd35_compile.png
```

--------------------------------

### Download SDXL Quantization Info for ComfyUI

Source: https://github.com/siliconflow/onediff/blob/main/README_ENTERPRISE.md

Use this command to download the SDXL quantization info file and place it in the ComfyUI models onediff_quant directory.

```bash
wget https://huggingface.co/siliconflow/sdxl-base-1.0-onediff-comfy-enterprise-v1/resolve/main/sd_xl_base_1.0_quantize_info.pt -O ComfyUI/models/onediff_quant/sd_xl_base_1.0_quantize_info.pt
```

--------------------------------

### Check OneFlow Doctor

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/OnlineQuantization.md

Run this command to check your OneFlow installation details, including version and Git commit.

```bash
python -m oneflow --doctor
```

--------------------------------

### Run Text-to-Image Benchmark with Quantization

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/pixart/README.md

Executes the text-to-image benchmark script with specific Nexfort compiler configurations and quantization settings.

```bash
python3 ./benchmarks/text_to_image.py \
--model PixArt-alpha/PixArt-XL-2-1024-MS \
--scheduler none \
--steps 20 \
--output-image ./pixart_alpha_fp8_90%.png \
--prompt "product photography, world of warcraft orc warrior, white background" \
--compiler nexfort \
--compiler-config '{"mode": "quant:max-optimize:max-autotune:freezing:benchmark:low-precision:cudagraphs", "memory_format": "channels_last"}' \
--quantize \
--quantize-config '{"quant_type": "fp8_e4m3_e4m3_dynamic"}' \
--quant-submodules-config-path /path/to/fp8_e4m3.json
```

--------------------------------

### Download SDXL Base Model

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/docs/ComfyUI_Online_Quantization.md

Use wget to download the SDXL base model safetensors file into the ComfyUI checkpoints directory. This is required for running SDXL workflows.

```bash
cd ComfyUI
wget -O models/checkpoints/sd_xl_base_1.0.safetensors https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors
```

--------------------------------

### Register a Workflow Generator

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/benchmarks/README.md

Example of using the @register_generator decorator to define a workflow generator for testing with specific image dimensions.

```python
# file: onediff/onediff_comfy_nodes/benchmarks/src/input_registration.py

@register_generator(f"{WORKFLOW_DIR}/example_workflow_api.json")
def _(workflow_path, *args, **kwargs):
    with open(workflow_path, "r") as fp:
        workflow = json.load(fp)
    graph = ComfyGraph(graph=workflow, sampler_nodes=["3"])
    for height in [1024, 768, 512]:
        for width in [1024, 768, 512]:
            graph.set_image_size(height=height, width=width)
            yield InputParams(graph=graph)
```

--------------------------------

### Run SD1.5 Benchmark Without Compilation

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd1.5/README.md

Execute the baseline text-to-image benchmark without using the compiler.

```shell
python3 benchmarks/text_to_image.py \
  --model runwayml/stable-diffusion-v1-5 \
  --height 512 --width 512 \
  --scheduler none \
  --steps 20 \
  --output-image ./stable-diffusion-v1-5.png \
  --prompt "beautiful scenery nature glass bottle landscape, , purple galaxy bottle," \
  --compiler none \
  --seed 1 \
  --print-output
```

--------------------------------

### Install ComfyUI Custom Nodes

Source: https://github.com/siliconflow/onediff/blob/main/onediff_comfy_nodes/modules/oneflow/hijack_comfyui_instantid/README.md

Clone the ComfyUI custom nodes repository and reset to a specific commit. Ensure you are in the 'ComfyUI/custom_nodes' directory before running.

```shell
cd ComfyUI/custom_nodes
git clone https://github.com/comfyanonymous/ComfyUI
git reset --hard 2d4164271634476627aae31fbec251ca748a0ae0
```

--------------------------------

### Run SD2 Baseline Benchmark

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd2/README.md

Execute the text-to-image benchmark without compilation to establish a baseline.

```shell
python3 benchmarks/text_to_image.py \
  --model stabilityai/stable-diffusion-2-1 \
  --height 768 --width 768 \
  --scheduler none \
  --steps 20 \
  --output-image ./stable-diffusion-2-1.png \
  --prompt "beautiful scenery nature glass bottle landscape, , purple galaxy bottle," \
  --compiler none \
  --print-output
```

--------------------------------

### Troubleshoot CUDA Error in NexFort Compilation

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd2/README.md

If you encounter a CUBLAS_STATUS_NOT_SUPPORTED error during NexFort compilation, adjust the --compiler-config by setting 'inductor.optimize_linear_epilogue' to false. This configuration is shown in the example command.

```text
torch._dynamo.exc.BackendCompilerFailed: backend='nexfort' raised:
ErrorFromChoice: CUDA error: CUBLAS_STATUS_NOT_SUPPORTED when calling `cublasLtMatmul( ltHandle, computeDesc.descriptor(), &alpha_val, mat1_ptr, Adesc.descriptor(), mat2_ptr, Bdesc.descriptor(), &beta_val, mat3_ptr, Ddesc.descriptor(), result_ptr, Cdesc.descriptor(), &heuristicResult.algo, workspace.data_ptr(), workspaceSize, at::cuda::getCurrentCUDAStream())`
From choice ExternKernelCaller(extern_kernels.nexfort_cuda_linear_epilogue)
inputs = [
    torch.empty_strided((2, 10, 4096, 64), (2621440, 64, 640, 1), dtype=torch.float16, device='cuda'),
    torch.empty_strided((640, 640), (640, 1), dtype=torch.float16, device='cuda'),
    torch.empty_strided((640,), (1,), dtype=torch.float16, device='cuda'),
    torch.empty_strided((2, 4096, 640), (2621440, 640, 1), dtype=torch.float16, device='cuda'),
]
out = torch.empty_strided((2, 4096, 640), (2621440, 640, 1), dtype=torch.float16, device='cuda')
```

--------------------------------

### Run SD2 with NexFort Compilation

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd2/README.md

Execute the text-to-image benchmark using the NexFort compiler with specific configuration options.

```shell
python3 benchmarks/text_to_image.py \
  --model stabilityai/stable-diffusion-2-1 \
  --height 768 --width 768 \
  --scheduler none \
  --steps 20 \
  --output-image ./stable-diffusion-2-1-compile.png \
  --prompt "beautiful scenery nature glass bottle landscape, , purple galaxy bottle," \
  --compiler nexfort \
  --compiler-config '{"mode": "cudagraphs:benchmark:max-autotune:low-precision:cache-all", "memory_format": "channels_last", "options": {"triton.fuse_attention_allow_fp16_reduction": false, "inductor.optimize_linear_epilogue": true, "overrides.conv_benchmark": true, "overrides.matmul_allow_tf32": true}}' \
  --print-output
```

--------------------------------

### Configure Compiler Modes

Source: https://github.com/siliconflow/onediff/blob/main/src/onediff/infer_compiler/README.md

Define optimization strategies using the options dictionary.

```python
options = {"mode": "max-optimize:max-autotune:low-precision:cache-all"}
```

```python
options = {"mode": "O3"}
```

--------------------------------

### Test PixArt Alpha with Dynamic Shape

Source: https://github.com/siliconflow/onediff/blob/main/src/onediff/infer_compiler/backends/nexfort/README.md

Runs the PixArt Alpha model with the nexfort backend, enabling dynamic shape inference. This example uses a specific model and scheduler configuration, along with `max-optimize:max-autotune:low-precision` mode and `channels_last` memory format.

```bash
python3 ./benchmarks/text_to_image.py \
--model PixArt-alpha/PixArt-XL-2-1024-MS \
--scheduler none \
--steps 20 \
--height 512 \
--width 768 \
--compiler nexfort \
--compiler-config '{"mode": "max-optimize:max-autotune:low-precision", "memory_format": "channels_last", "dynamic": true}' \
--run_multiple_resolutions 1
```

--------------------------------

### Run SDXL With NexFort Compilation

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sdxl/README.md

Execute the text-to-image benchmark script for SDXL with the NexFort compiler enabled. This command includes specific compiler configurations for optimization.

```shell
python3 benchmarks/text_to_image.py \
  --model stabilityai/stable-diffusion-xl-base-1.0 \
  --height 1024 --width 1024 \
  --scheduler none \
  --steps 20 \
  --output-image ./stable-diffusion-xl-compile.png \
  --prompt "beautiful scenery nature glass bottle landscape, , purple galaxy bottle," \
  --compiler nexfort \
  --compiler-config '{"mode": "benchmark:cudagraphs:max-autotune:low-precision:cache-all", "memory_format": "channels_last", "options": {"inductor.optimize_linear_epilogue": true, "overrides.conv_benchmark": true, "overrides.matmul_allow_tf32": true}}' \
  --variant fp16 \
  --seed 1 \
  --print-output
```

--------------------------------

### Run Text-to-Image with SD3 and Dynamic Shape

Source: https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/sd3/README.md

Execute the text-to-image script for Stable Diffusion 3 using OneDiff. This example utilizes a specific compiler configuration for dynamic shapes, enabling `max-optimize`, `max-autotune`, `low-precision`, and `cache-all` modes. It also specifies memory format and enables dynamic resolution.

```bash
python3 onediff_diffusers_extensions/examples/sd3/text_to_image_sd3.py     --compiler-config '{"mode": "max-optimize:max-autotune:low-precision:cache-all", "memory_format": "channels_last", "dynamic": true}'     --height 512     --width 768     --run_multiple_resolutions 1     --saved-image sd3_compile.png
```