Conditionally installing hardware-accelerated PyTorch with Poetry on different hardware using the same pyproject.toml can be tricky. This repo serves as a quick lookup for the configuration file and installation commands.
| Command | Behavior |
|---|---|
poetry install --sync |
Does not install PyTorch (import fails). |
poetry install --sync -E cpu |
Installs PyTorch with CPU only. |
poetry install --sync -E cuda --with cuda |
Installs the CUDA variant of PyTorch. Expects NVIDIA hardware. |
Note
The --sync flag is optional, but useful for trying out both CPU and CUDA modes on a same machine/container, or if packages were removed. This will make sure packages that are only in the previous configuration are also removed from your current environment.
Warning
The example below is likely not what you want:
| Command | Behavior |
|---|---|
poetry install --sync -E cuda |
Actually installs the CPU variant of PyTorch without errors or warnings. |
if lspci | grep -i nvidia; then
poetry install --sync --extras=cuda --with cuda
else
poetry install --sync --extras=cpu
fipoetry run python check-cuda.pyor
poetry run python -c "import torch; print(torch.cuda.is_available())"