安装Yolo_Nas时执行pip install super_gradients遭遇依赖冲突

问题背景 #

安装YOLO-NAS时执行pip install super_gradients触发依赖冲突。当前环境已安装torch 2.0.0，但super_gradients各版本的依赖要求与该版本不兼容：

• 3.1.x版本要求torch >=1.9.0 且 <1.14
• 更早版本依赖onnxruntime，与现有环境存在冲突

执行的命令 #

cd C:\Users\Isaac
pip install super_gradients

错误日志 #

Collecting super_gradients
  Using cached super_gradients-3.1.1-py3-none-any.whl (964 kB)
INFO: pip is looking at multiple versions of super-gradients to determine which version is compatible with other requirements. This could take a while.
  Using cached super_gradients-3.1.0-py3-none-any.whl (965 kB)
  Using cached super_gradients-3.0.9-py3-none-any.whl (938 kB)
  Using cached super_gradients-3.0.8-py3-none-any.whl (892 kB)
  Using cached super_gradients-3.0.7-py3-none-any.whl (794 kB)
  Using cached super_gradients-3.0.6-py3-none-any.whl (762 kB)
  Using cached super_gradients-3.0.5-py3-none-any.whl (748 kB)
  Using cached super_gradients-3.0.4-py3-none-any.whl (748 kB)
INFO: pip is looking at multiple versions of super-gradients to determine which version is compatible with other requirements. This could take a while.
  Using cached super_gradients-3.0.3-py3-none-any.whl (732 kB)
Requirement already satisfied: torch>=1.9.0 in c:\python311\lib\site-packages (from super_gradients) (2.0.0)
Requirement already satisfied: tqdm>=4.57.0 in c:\python311\lib\site-packages (from super_gradients) (4.65.0)
Collecting boto3>=1.17.15 (from super_gradients)
  Using cached boto3-1.26.126-py3-none-any.whl (135 kB)
Collecting jsonschema>=3.2.0 (from super_gradients)
  Using cached jsonschema-4.17.3-py3-none-any.whl (90 kB)
Collecting Deprecated>=1.2.11 (from super_gradients)
  Using cached Deprecated-1.2.13-py2.py3-none-any.whl (9.6 kB)
Requirement already satisfied: opencv-python>=4.5.1 in c:\python311\lib\site-packages (from super_gradients) (4.7.0.72)
Requirement already satisfied: scipy>=1.6.1 in c:\python311\lib\site-packages (from super_gradients) (1.10.1)
Requirement already satisfied: matplotlib>=3.3.4 in c:\python311\lib\site-packages (from super_gradients) (3.7.1)
Requirement already satisfied: psutil>=5.8.0 in c:\python311\lib\site-packages (from super_gradients) (5.9.5)
Collecting tensorboard>=2.4.1 (from super_gradients)
  Using cached tensorboard-2.12.3-py3-none-any.whl (5.6 MB)
Requirement already satisfied: setuptools>=21.0.0 in c:\python311\lib\site-packages (from super_gradients) (65.5.0)
Collecting coverage~=5.3.1 (from super_gradients)
  Using cached coverage-5.3.1.tar.gz (684 kB)
  Preparing metadata (setup.py) ... done
Requirement already satisfied: torchvision>=0.10.0 in c:\python311\lib\site-packages (from super_gradients) (0.15.1)
Collecting sphinx~=4.0.2 (from super_gradients)
  Using cached Sphinx-4.0.3-py3-none-any.whl (2.9 MB)
Collecting sphinx-rtd-theme (from super_gradients)
  Using cached sphinx_rtd_theme-1.2.0-py2.py3-none-any.whl (2.8 MB)
Collecting torchmetrics==0.8 (from super_gradients)
  Using cached torchmetrics-0.8.0-py3-none-any.whl (408 kB)
Requirement already satisfied: pillow>=9.2.0 in c:\python311\lib\site-packages (from super_gradients) (9.5.0)
Collecting hydra-core>=1.2.0 (from super_gradients)
  Using cached hydra_core-1.3.2-py3-none-any.whl (154 kB)
Collecting omegaconf (from super_gradients)
  Using cached omegaconf-2.3.0-py3-none-any.whl (79 kB)
Collecting super_gradients
  Using cached super_gradients-3.0.2-py3-none-any.whl (664 kB)
  Using cached super_gradients-3.0.1-py3-none-any.whl (635 kB)
  Using cached super_gradients-3.0.0-py3-none-any.whl (615 kB)
  Using cached super_gradients-2.6.0-py3-none-any.whl (11.0 MB)
INFO: This is taking longer than usual. You might need to provide the dependency resolver with stricter constraints to reduce runtime. See https://pip.pypa.io/warnings/backtracking for guidance. If you want to abort this run, press Ctrl + C.
  Using cached super_gradients-2.5.0-py3-none-any.whl (11.0 MB)
  Using cached super_gradients-2.2.0-py3-none-any.whl (10.9 MB)
  Using cached super_gradients-2.1.0-py3-none-any.whl (23.0 MB)
Collecting elasticsearch==7.15.2 (from super_gradients)
  Using cached elasticsearch-7.15.2-py2.py3-none-any.whl (379 kB)
Collecting CMRESHandler>=1.0.0 (from super_gradients)
  Using cached CMRESHandler-1.0.0-py2.py3-none-any.whl (15 kB)
Collecting super_gradients
  Using cached super_gradients-2.0.1-py3-none-any.whl (19.4 MB)
  Using cached super_gradients-2.0.0-py3-none-any.whl (19.4 MB)
  Using cached super_gradients-1.7.5-py3-none-any.whl (19.3 MB)
Collecting torchmetrics==0.7.3 (from super_gradients)
  Using cached torchmetrics-0.7.3-py3-none-any.whl (398 kB)
Collecting super_gradients
  Using cached super_gradients-1.7.4-py3-none-any.whl (19.3 MB)
  Using cached super_gradients-1.7.3-py3-none-any.whl (19.3 MB)
Collecting torchmetrics>=0.5.0 (from super_gradients)
  Using cached torchmetrics-0.11.4-py3-none-any.whl (519 kB)
Collecting super_gradients
  Using cached super_gradients-1.7.2-py3-none-any.whl (19.3 MB)
  Using cached super_gradients-1.7.1-py3-none-any.whl (15.1 MB)
  Using cached super_gradients-1.6.0-py3-none-any.whl (547 kB)
  Using cached super_gradients-1.5.2-py3-none-any.whl (540 kB)
  Using cached super_gradients-1.5.1-py3-none-any.whl (540 kB)
  Using cached super_gradients-1.5.0-py3-none-any.whl (497 kB)
  Using cached super_gradients-1.4.0-py3-none-any.whl (419 kB)
  Using cached super_gradients-1.3.1-py3-none-any.whl (416 kB)
  Using cached super_gradients-1.3.0-py3-none-any.whl (415 kB)
ERROR: Cannot install super-gradients==1.3.0, super-gradients==1.3.1, super-gradients==1.4.0, super-gradients==1.5.0, super-gradients==1.5.1, super-gradients==1.5.2, super-gradients==1.6.0, super-gradients==1.7.1, super-gradients==1.7.2, super-gradients==1.7.3, super-gradients==1.7.4, super-gradients==1.7.5, super-gradients==2.0.0, super-gradients==2.0.1, super-gradients==2.1.0, super-gradients==2.2.0, super-gradients==2.5.0, super-gradients==2.6.0, super-gradients==3.0.0, super-gradients==3.0.1, super-gradients==3.0.2, super-gradients==3.0.3, super-gradients==3.0.4, super-gradients==3.0.5, super-gradients==3.0.6, super-gradients==3.0.7, super-gradients==3.0.8, super-gradients==3.0.9, super-gradients==3.1.0 and super-gradients==3.1.1 because these package versions have conflicting dependencies.

The conflict is caused by:
    super-gradients 3.1.1 depends on torch<1.14 and >=1.9.0
    super-gradients 3.1.0 depends on torch<1.14 and >=1.9.0
    super-gradients 3.0.9 depends on torch<1.14 and >=1.9.0
    super-gradients 3.0.8 depends on torch<1.14 and >=1.9.0
    super-gradients 3.0.7 depends on torch<1.14 and >=1.9.0
    super-gradients 3.0.6 depends on torch<=1.12 and >=1.9.0
    super-gradients 3.0.5 depends on torch<=1.12 and >=1.9.0
    super-gradients 3.0.4 depends on torch<=1.12 and >=1.9.0
    super-gradients 3.0.3 depends on onnxruntime
    super-gradients 3.0.2 depends on onnxruntime
    super-gradients 3.0.1 depends on onnxruntime
    super-gradients 3.0.0 depends on onnxruntime
    super-gradients 2.6.0 depends on onnxruntime
    super-gradients 2.5.0 depends on onnxruntime
    super-gradients 2.2.0 depends on onnxruntime
    super-gradients 2.1.0 depends on onnxruntime
    super-gradients 2.0.1 depends on onnxruntime
    super-gradients 2.0.0 depends on onnxruntime
    super-gradients 1.7.5 depends on onnxruntime
    super-gradients 1.7.4 depends on onnxruntime
    super-gradients 1.7.3 depends on onnxruntime
    super-gradients 1.7.2 depends on onnxruntime
    super-gradients 1.7.1 depends on onnxruntime
    super-gradients 1.6.0 depends on onnxruntime
    super-gradients 1.5.2 depends on onnxruntime
    super-gradients 1.5.1 depends on onnxruntime
    super-gradients 1.5.0 depends on onnxruntime
    super-gradients 1.4.0 depends on onnxruntime
    super-gradients 1.3.1 depends on onnxruntime
    super-gradients 1.3.0 depends on onnxruntime

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts

解决方法 #

1. 降级Torch到兼容版本 #

super_gradients 3.1.x版本要求Torch版本在1.9.0~1.14之间，可卸载现有Torch后安装指定版本：

pip uninstall torch torchvision -y
# 以CUDA 11.7为例，安装Torch 1.13.1
pip install torch==1.13.1 torchvision==0.14.1 --index-url https://download.pytorch.org/whl/cu117
# 再安装super_gradients
pip install super_gradients

若使用CPU版本，可执行：

pip install torch==1.13.1 torchvision==0.14.1 --index-url https://download.pytorch.org/whl/cpu

2. 使用虚拟环境隔离依赖 #

创建独立虚拟环境，避免与现有环境依赖冲突，pip会自动安装兼容的Torch版本：

# 使用venv创建虚拟环境（Python内置）
python -m venv yolonas_env
# Windows激活环境
yolonas_env\Scripts\activate
# Linux/macOS激活环境
source yolonas_env/bin/activate
# 安装super_gradients
pip install super_gradients

3. 强制安装（不推荐） #

若仅需临时测试，可忽略依赖冲突强制安装，但可能导致运行时错误：

pip install super_gradients --force-reinstall --no-deps

内容的提问来源于stack exchange，提问作者isaak mwangi