How to convert pip install to Poetry file?

Question

With hours of research, I still can't figure out a way to convert this pip install cmd to pyproject.toml file. I'm trying to install PyTorch.

pip install torch==1.7.0+cpu torchvision==0.8.1+cpu torchaudio===0.7.0 -f https://download.pytorch.org/whl/torch_stable.html

This is what I got at the moment (Completely Wrong!)

[tool.poetry]
name = "poetry-test"
version = "0.1.0"
description = ""
authors = ["Your Name <you@example.com>"]

[tool.poetry.dependencies]
python = "^3.8"
torch = "^1.7.0"
torchvision = "^0.8.1"
torchaudio = "^0.7.0"

[tool.poetry.dependencies.torch]
url = "https://download.pytorch.org/whl/torch_stable.html"

[tool.poetry.dev-dependencies]

[build-system]
requires = ["poetry>=0.12"]
build-backend = "poetry.masonry.api"

Answer 1

If you use a url specifier in your dependency definition, it needs to point directly to the file that you want to install, and not an .html page:

[tool.poetry.dependencies]
python = "^3.8"
torch = {url = "https://download.pytorch.org/whl/cpu/torch-1.7.0%2Bcpu-cp38-cp38-win_amd64.whl"}
torchaudio = {url = "https://download.pytorch.org/whl/torchaudio-0.7.0-cp38-none-win_amd64.whl"}
torchvision = {url = "https://download.pytorch.org/whl/cpu/torchvision-0.8.1%2Bcpu-cp38-cp38-win_amd64.whl"}

That being said, your setup can't be installed with poetry right now because you're going to get a solver error:

[SolverProblemError]
Because torchaudio (0.7.0) depends on torch (1.7.0) which doesn't match any versions, torchaudio is forbidden.
So, because no versions of torchaudio match !=0.7.0
 and poetry-test depends on torchaudio (*), version solving failed.

While torch can be installed in isolation, something goes wrong while matching versions when installing torchaudio or torchvision . Maybe it's got to do with the fact that torch includes build info in their version in the wheel you're trying to install (ie 1.7.0+cpu ), maybe it's because url-denoted dependencies can't be assigned a version in the pyproject.toml (which makes sense, it's not like poetry gets to resolve anything if we tell it "here is a tarball, install it!"), or maybe just an internal poetry bug.

You might have to stick with pip for now, given that it will not complain about solving a dependency tree and instead just starts installing. But ultimately, blame the torch maintainers, for 1) not uploading all their wheels to the global Python package index , and 2) not even setting up their own file server in a PEP-503 conform way.

---

Update:

I looked a bit more into the discussions, and it seems that a helpful someone has done torch 's job of serving their wheels on a PEP-503 conform server. So for the time being, the following will actually work:

[tool.poetry]
name = "poetry-test"
version = "0.1.0"
description = ""
authors = ["Your Name <you@example.com>"]

[[tool.poetry.source]]
name = "foo"
url = "https://vxlabs.com/pypi//simple/"

[tool.poetry.dependencies]
python = "^3.8"
torch = "1.7.0"
torchvision = "0.8.1"
torchaudio = "0.7.0"

[build-system]
requires = ["poetry>=0.12"]
build-backend = "poetry.masonry.api"

Answer 2

Previous solution didnt work for me as mentioned in this poetry issue#2543 . So what worked for me in the meantime was to upgrade to version 1.2(preview) which addresses that issue.

Install poetry 1.2

curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/install-poetry.py | python - --preview

Add below repo to your pyproject.toml as suggested here issue#4124 , and you shouldnt get version mismatch.

[[tool.poetry.source]]
name = "torch_rep"
url = "https://eternalphane.github.io/pytorch-pypi"

After that, you can just proceed to install them poetry add torch =1.7.0+cpu poetry add torchvision=0.8.1+cpu

Alternatively, adding dependencies to pyproject.toml and doing poetry install should also work.