Instructions for installing PyTorch #6409

davidgilbertson · 2022-09-05T08:59:06Z

I have searched the issues of this repo and believe that this is not a duplicate.

Issue

As mentioned in issue #4231 there is some confusion around installing PyTorch with CUDA but it is now somewhat resolved. It still requires a few steps, and all options have pretty serious flaws. Below are two options that 'worked' for me, on Poetry version 1.2.0.

Option 1 - wheel URLs for a specific platform

You will need to pick the specific wheels you want. These are listed here: https://download.pytorch.org/whl/torch_stable.html. E.g. if you want CUDA 11.6, Python 3.10 and Windows, search that page for cu116-cp310-cp310-win_amd64.whl to see the matches for torch, torchaudio and torchvision
In your pyproject.toml file add the URLs like:

[tool.poetry.dependencies]
python = "^3.10"
numpy = "^1.23.2"
torch = { url = "https://download.pytorch.org/whl/cu116/torch-1.12.1%2Bcu116-cp310-cp310-win_amd64.whl"}
torchaudio = { url = "https://download.pytorch.org/whl/cu116/torchaudio-0.12.1%2Bcu116-cp310-cp310-win_amd64.whl"}
torchvision = { url = "https://download.pytorch.org/whl/cu116/torchvision-0.13.1%2Bcu116-cp310-cp310-win_amd64.whl"}

Run poetry update. It will download a lot of data (many GB) and take quite some time. And this doesn't seem to cache reliably (at least, I've waited 30 minutes+ at 56 Mbps three separate times while troubleshooting this, for the exact same wheels)

Note that each subsequent poetry update will do another huge download and you'll see this message:

  • Updating torch (1.12.1+cu116 -> 1.12.1+cu116 https://download.pytorch.org/whl/cu116/torch-1.12.1%2Bcu116-cp310-cp310-win_amd64.whl)
  • Updating torchaudio (0.12.1+cu116 -> 0.12.1+cu116 https://download.pytorch.org/whl/cu116/torchaudio-0.12.1%2Bcu116-cp310-cp310-win_amd64.whl)
  • Updating torchvision (0.13.1+cu116 -> 0.13.1+cu116 https://download.pytorch.org/whl/cu116/torchvision-0.13.1%2Bcu116-cp310-cp310-win_amd64.whl)

Option 2 - alternate source

[tool.poetry.dependencies]
python = "^3.10"
numpy = "^1.23.2"
torch = { version = "1.12.1", source="torch"}
torchaudio = { version = "0.12.1", source="torch"}
torchvision = { version = "0.13.1", source="torch"}

[[tool.poetry.source]]
name = "torch"
url = "https://download.pytorch.org/whl/cu116"
secondary = true

This seems to have worked (although I already had the packages installed) but it reports errors like Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pillow/, but I think they get installed anyway (maybe a better message would be "Can't access pillow at 'https://download.pytorch.org/whl/cu116', falling back to pypi")

Also, if you later go on to do, say poetry add pandas (a completely unrelated library) you'll get a wall of messages like:

Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pandas/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pandas/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pytz/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/python-dateutil/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/numpy/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pillow/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/requests/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/typing-extensions/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/certifi/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/urllib3/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/idna/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/charset-normalizer/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/python-dateutil/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/six/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/pytz/
Source (torch): Authorization error accessing https://download.pytorch.org/whl/cu116/six/

This happens with or without secondary = true in the source config.

Maintainers: please feel free to edit the text of this if I've got something wrong.

The text was updated successfully, but these errors were encountered:

dimbleby · 2022-09-05T09:20:09Z

Failure to cache during resolution is covered by #2415, poetry's insistence on checking all sources for all packages is discussed at #5984

variolam · 2022-09-06T12:31:54Z

Solution 1 seems infeasible when working in a team with machines on different operating systems, due to the need of providing the complete URL of the wheel including operating system and exact version number.

Solution 2 seems to work, but it results in downloading every single PyTorch version that can found, independent on the operating system. I'm running Windows and the download looks like this:

A single installation takes around 15-20 minutes at ~250 Mbps.

neersighted · 2022-09-06T13:17:03Z

Poetry will always download wheels for every platform when you install -- this is because there is no other way to get package metadata from a repository using PEP 503's API.

Queuecumber · 2022-09-06T14:28:43Z

Can you elaborate a little on what metadata is needed and why downloading every conceivable version of a package yields that metadata? As mentioned this leads to a ~20min install for one package

radoering · 2022-09-06T14:37:59Z

Solution 1 seems infeasible when working in a team with machines on different operating systems, due to the need of providing the complete URL of the wheel including operating system and exact version number.

It's not convenient, but it should be feasible with multiple constraints dependencies.

neersighted · 2022-09-06T14:43:07Z

Poetry requires the package's core metadata aka the METADATA file (most critically this includes dependencies), as well as the bdist/sdist itself for hashing purposes. Note that PEP 658 is a standard for serving the METADATA file that is implementable by third-party repositories, and PEP 691 specifies a (potentially) richer JSON API (including hashes) that third-party repositories could likely implement.

However, Poetry is unlikely to grow support for these new APIs until PyPI does, and I think third party repos are unlikely to implement it before PyPI. Eventually support for these APIs will allow for feature and performance parity in Poetry between PyPI and third-party repositories. Until then, we are stuck with the legacy HTML API, which requires us to download every package when generating a lock file for the first time.

After your cache is warm you will not need to download again, and on other platforms you will only download the necessary files as the metadata is captured in the lock file.

Queuecumber · 2022-09-06T14:46:37Z

What I'm not understanding is that poetry knows I'm on Linux with python 3.8 but it still downloads

https://download.pytorch.org/whl/cu116/torch-1.12.1%2Bcu116-cp37-cp37-win_amd64.whl

Or does that wheel not contain the core metadata that is needed?

Queuecumber · 2022-09-06T14:52:23Z

Also there seems to be a second problem going on here unless I've misunderstood the documentation

I have this in my pyproject.toml

[[tool.poetry.source]]
name = "torchcu116"
url = "https://download.pytorch.org/whl/cu116"
default = false
secondary = true

i.e. secondary = true

Yet poetry is asking that repository for every package I try to install. I thought from the documentation that secondary meant it would go to pypi for any package unless specifically asked to go to that custom repository.

neersighted · 2022-09-06T14:59:19Z

What I'm not understanding is that poetry knows I'm on Linux with python 3.8 but it still downloads
https://download.pytorch.org/whl/cu116/torch-1.12.1%2Bcu116-cp37-cp37-win_amd64.whl
Or does that wheel not contain the core metadata that is needed?

Poetry constructs a universal lock file -- we write hashes to the lock file for all supported platforms. Thus on the first machine you generate a lock file, you will download a wheel for every supported platform. There is no way to write hashes to the lock file for those foreign/other platform versions without downloading them first.

If you want to reduce the scope of this a bit, you can tighten your Python constraint. There is a prototype of a new feature at #4956 (though it needs resurrection, design, and testing work) to add arbitrary markers to let a project reduce its supported platforms as an opt-in.

Also there seems to be a second problem going on here unless I've misunderstood the documentation

I have this in my pyproject.toml
[[tool.poetry.source]]
name = "torchcu116"
url = "https://download.pytorch.org/whl/cu116"
default = false
secondary = true
i.e. secondary = true

Yet poetry is asking that repository for every package I try to install. I thought from the documentation that secondary meant it would go to pypi for any package unless specifically asked to go to that custom repository.

I think it might be you misreading -- that is the intended and documented behavior. There is a proposal to introduce new repository types at #5984 (comment) as the current secondary behavior covers multiple use cases poorly, while being ideal for none of them.

Queuecumber · 2022-09-06T15:42:46Z

Poetry constructs a universal lock file

OK this makes sense now, thanks for the explanation, looking forward to that PR hopefully being merged eventually

I think it might be you misreading

I was misreading I see that this is intended behavior

the current secondary behavior covers multiple use cases poorly

I agree with that and I hope that these new repository types can be implemented

ZetiMente · 2022-09-24T23:49:36Z

Can't wait for option 2 to have good performance !

neersighted · 2022-09-25T00:22:33Z

Please 👍 on issues instead of commenting me too -- it keeps the notifications down and still shows interest. Thanks!

Poetry is not yet ready to handle the different versions of PyTorch and Torchvision. See the related issue: python-poetry/poetry#6409

voegtlel · 2022-10-12T10:59:09Z

Probably a follow-up issue on the second option:
If I do poetry install on that, I get

Installing dependencies from lock file

Package operations: 50 installs, 0 updates, 0 removals

  • Installing wrapt (1.12.1): Failed

  KeyringLocked

  Failed to unlock the collection!

  at ~/.local/share/pypoetry/venv/lib/python3.10/site-packages/keyring/backends/SecretService.py:67 in get_preferred_collection
       63│             raise InitError("Failed to create the collection: %s." % e)
       64│         if collection.is_locked():
       65│             collection.unlock()
       66│             if collection.is_locked():  # User dismissed the prompt
    →  67│                 raise KeyringLocked("Failed to unlock the collection!")
       68│         return collection
       69│ 
       70│     def unlock(self, item):
       71│         if hasattr(item, 'unlock'):

I guess it tries to load that from the secondary repo as well, and expects to use the keyring due to the unauthorized thing?

neersighted · 2022-10-12T12:13:21Z

That's #1917 -- our use of keyring hits surprisingly many system configurations in which hard errors occur, and it needs some work.

felix-ht · 2022-10-18T13:13:00Z

@neersighted at least PyPI seems have started work on the new json APi

neersighted · 2022-10-18T13:15:36Z

Indeed, Poetry 1.2.2 relies on the new PEP 691 support. However, PEP 658 is the real blocker for better performance in third-party repos -- there is a long-running PR blocked on review and a rather combative contributor, but otherwise no major progress on that front.

MaKaNu · 2022-10-20T11:54:03Z

could you add a link to the blocked PR? I switched today to method 1 because method 2 took ages. Seems the meta servers are slowed down today... Dependency resolving which takes up to 4000 seconds for method 1 is also insane. And then it failed because I accidental copied 1.12.0 instead of 1.12.1 for the windows release. I really like the idea of poetry but this needs huge improvement.

felix-ht · 2022-10-20T12:43:40Z

We use neither these approaches.
My hacky solution to just install the base version of pytorch say 1.12.1 with poetry and then install the specific versions needed for the machines with pip in a makefile command.

As the gpu versions have the same dependencies as the base version this should be OK.

The big downsides are

download both torch versions
lockfile does not mach installed version (of pytorch)
a change in the pytorch version requires and update to the makefile as well as the pyproject.toml
any poetry action like poetry add will reinstall the base version
breaks Packaging - usually not a concern for ml projects

The big upside is that it is very easy to create make scripts for different machines - and that its pretty fast (very important for ci/cd)
Example:

[tool.poetry.dependencies]
python = "^3.10"
numpy = "^1.23.2"
torch = "1.12.1"
torchvision = "0.13.1"

install_cu116:
	poetry install
	poetry run pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 -f https://download.pytorch.org/whl/torch_stable.html

timothyjlaurent · 2022-11-03T03:37:34Z

install_cu116:
	poetry install
	poetry run pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 -f https://download.pytorch.org/whl/torch_stable.html

this has essentially been our approach, but in Dockerfiles, gated by a build arg

ARG TORCH_ARCH="cpu"
#ARG TORCH_ARCH="cu113"

RUN  poetry install -vvv --no-root --no-dev \
    && pip install -U wheel torch==1.12.1+${TORCH_ARCH} torchvision==0.13.1+${TORCH_ARCH} -f https://download.pytorch.org/whl/torch_stable.html \
    && pip uninstall poetry -y \
    && rm -rf ~/.config/pypoetry \
    && rm -rf /root/.cache/pip

This also allows installing the CPU version which is smaller a smaller package that lacks the CUDA drivers that come in the normal pytorch package from pypi. They help to slim the images down.

brochier · 2022-11-10T11:55:04Z

Hi ! Do you know if it's possible to specify two different optional versions of torch in the pyproject.toml ? I would like to use a cpu version locally and a gpu version on a distant server. You can have a look at an example in this stack overflow post.

neersighted · 2022-11-10T13:21:07Z

That is something not unlike #5222; the consensus has been that as Poetry is an interoperable tool, no functionality will be added to the core project to support this until there is a standards-based method. A plugin can certainly support this with some creativity and would be the immediate "I want Poetry to support this" use case solution in my mind.

robinbrochier · 2022-11-10T13:41:24Z

#5222

Thanks for your answer, do you have any such plugin in mind ?

neersighted · 2022-11-10T13:46:59Z

I don't have any links at hand, but building on top of Light the Torch has been discussed. But if you mean if I know anyone is working on one, no, not that I am aware of.

arthur-st · 2022-11-10T16:39:11Z

For as long as pip install torch does Just Work ™️, having to put any effort at all into installing torch via poetry will be a vote against using it in commercial ML work that requires torch or similar hardware-bound components. Assuming that one is open to consider poetry, conda, and pipenv as functionally equivalent options, of course.

neersighted · 2022-11-10T16:44:22Z

I don't see that as a Poetry problem -- the issue is that there is an unmet packaging need, and no one from the ML world is working with the PyPA to define how to handle this robustly/no one from the PyPA has an interest in solving it.

Likewise, there is no interest in Poetry in supporting idiosyncratic workarounds for a non-standard and marginally compatible ecosystem; we'll be happy to implement whatever standards-based process evolves to handle these binary packages, but in the mean time any special-casing and package-specific functionality belong in a plugin and not Poetry itself.

PyroGenesis · 2024-05-15T18:06:13Z

What is possible regarding torch & poetry

Install torch from torch's repository: use the source flag, see original post, Option 2: alternate source

Use any single hardware-flavor of torch: use the source flag, see original post, Option 2: alternate source

Note that the secondary priority for sources is deprecated now. Personally I use the explicit priority to make sure it directly checks the pytorch index for torch rather than using it as a fallback.

[[tool.poetry.source]]
name = "pytorch-cuda"
url = "https://download.pytorch.org/whl/cu121"
priority = "explicit"

[tool.poetry.dependencies]
python = "~3.10"
torch = { version = "2.1.1+cu121", source = "pytorch-cuda" }

Also, another issue many users had was that poetry would download every wheel of a particular pytorch version to compute the hashes. The solution we currently have is to remind the pytorch devs to add hashes to their wheels if the index regresses again.

slashtechno · 2024-05-15T21:39:14Z

It will make a non-functional wheel/sdist which depends on torch but does not declare that dependency, indeed. Group only work in your local, poetry-management development environment and are not translated to the wheel/sdist. You cannot use groups with poetry build

You're right that the summary doesn't mention that either, it is an important caveat. Added that to the summary

Can the devs implement a fix for this? Or is this by-design?

QuentinSoubeyranAqemia · 2024-05-16T08:29:27Z

@slashtechno It's by design. The poetry build can only leverage what the python packaging specification permits, since its building wheel/sdist for tools other than poetry (e.g. pip). Groups are not part of the PyPA specification (only extras) and won't be understood by those downstream tools.

@PyroGenesis Thank you, I added a note regarding supplemental/priority in the summary. I missed that the example was so old.

david-waterworth · 2024-05-28T22:43:00Z

I don't know if it's already been proposed, but I'd like to see some sort of "build target" option (specifically for the non-package mode) which enabled some way of specifying mutually exclusive versions of the same package/group of packages (i.e. torch cpu/cuda/rocm, faiss cpu/gpu etc) based on some target flag.

It's a fairly common use case in ML, you often see multiple requirement.txt files depending on the users needs. A challenge is probably that downstream libraries like huggingface transformers require torch and that requirement can be met by any of the different torch wheels (i.e. cpu, cuda).

Without this I find myself getting painted into a corner with poetry (I know work 50:50 between a linux workstation with gpu, and an M3 MacBook and at the moment have to hack together a combination of installing some stuff with pip and the rest with poetry)

QuentinSoubeyranAqemia · 2024-05-29T10:02:25Z

@david-waterworth
For non-package mode, that looks like an additional feature for poetry's custom dependency groups when they are optional, to mark them as incompatible with another. I'd open a new issue/feature request and link to this discussion.

From my own testing, I think you can achieve this effect by marking all dependencies in a group with the extra != "other-group" environment marker, but I'm not quite sure, things are complicated at that point.

Now for your particular use-case of linux workstation and MacBook, this is already possible without the need for any flag: use environments markers (poetry doc), in particular the platform_system one.
Poetry lets you write this more easily by writing your dependency as a TOML object, see this documentation example which uses platform and source like you want in the second example.

As an additional note, I think the PyPI's version of torch does work when cuda is not available, it just runs on the cpu.

david-waterworth · 2024-05-30T03:18:28Z

@QuentinSoubeyranAqemia I've not had any luck with dependency groups and extra's - i.e. I get the same as #6409 (comment)

Thanks for the environment markers pointer, that should be useful. I'd still like to be able to conditionally install the CPU or GPU verison on linux as well though, I quite often have to switch between CPU or GPU docker images so most of my projects now use poetry for the base sagemaker/boto dependencies but the containers I build for ML pipelines all use requirements.txt

QuentinSoubeyranAqemia · 2024-05-30T10:08:12Z

Thanks for the environment markers pointer, that should be useful. I'd still like to be able to conditionally install the CPU or GPU verison on linux as well though

As detailed in the summary, this is currently not possible. This is a limitation of the python's packaging spec, it's not inherent to poetry. The best you can do to my knowledge is switch based on your platform.

Note that when exporting poetry-managed to requirements.txt using poetry-export, you can select poetry dependency groups as you please with --with and --without

david-waterworth · 2024-05-30T10:40:39Z

@QuentinSoubeyranAqemia that’s where I get frustrated with poetry - on the home page it’s advertised as “PYTHON PACKAGING AND DEPENDENCY MANAGEMENT MADE EASY” but in reality everything I’ve read in here indicates the focus is more or less only on packaging and the associated standards compliance - there’s a use case for dependency management without packaging which hopefully will see improved support over time.

I need to take a closer look at ‘poetry-export’ - it could be useful

tfha · 2024-06-12T13:54:13Z

A well functioning solution for installing pytorch with cuda using poetry.
I have battled with this problem for a while as well. I just want to post the solution I have used, that seems to be stable without too many warnings, and which works when you install other dependencies, in a complex training session etc. Its basically similar to what @PyroGenesis has described. I just want to add the poetry commands and some extra info as well.

I use Ubuntu (WSL) and Python version 3.12. You find the url for the right wheels file from pytorch getting started. Here I use the cuda 12.1 version.

You first change the source for where poetry tries to find wheels for pytorch with:

poetry source add --priority=explicit pytorch https://download.pytorch.org/whl/cu121

Your pyproject.toml file will then be updated with this:

[tool.poetry.dependencies]
python = "^3.12"
torch = {version = "^2.3.1+cu121", source = "pytorch"}

[[tool.poetry.source]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
priority = "explicit"

It is superimportant that the priority is set to explicit. This means that poetry will look for wheels in this source only if is requested explicitly by the name of the source, which I have set to "pytorch". All other packages will be installed from the standard pypi source, when you run poetry add <package> or poetry install.

You can then continue to install PyTorch with CUDA support by running the following command in the terminal:

poetry add torch --source pytorch

shuvamg007 · 2024-06-12T14:21:06Z

@QuentinSoubeyranAqemia that’s where I get frustrated with poetry - on the home page it’s advertised as “PYTHON PACKAGING AND DEPENDENCY MANAGEMENT MADE EASY” but in reality everything I’ve read in here indicates the focus is more or less only on packaging and the associated standards compliance - there’s a use case for dependency management without packaging which hopefully will see improved support over time.

I need to take a closer look at ‘poetry-export’ - it could be useful

@david-waterworth I have a similar use case wherein I use it for both packaging and dependency management. So while packaging doesn't work right now, it works pretty good for dependency management. I use the following config for example:

[tool.poetry]
name="<proj_name>"
...
readme="README.md"
package-mode=false

[tool.poetry.dependencies]
python = "~3.12"

[tool.poetry.group.cpu]
optional=true

[tool.poetry.group.cpu.dependencies]
torch={ version="~2.2", source="pytorch-cpu", markers="extra=='cpu' and extra!='cuda'" }
torchvision={ version="~0.17", source="pytorch-cpu", markers="extra=='cpu' and extra!='cuda'" }

[tool.poetry.group.cuda]
optional=true

[tool.poetry.group.cuda.dependencies]
torch={ version="~2.2", source="pytorch-cuda", markers="extra=='cuda' and extra!='cpu'" }
torchvision={ version="~0.17", source="pytorch-cuda", markers="extra=='cuda' and extra!='cpu'" }

[tool.poetry.extras]
cuda=["torch", "torchvision"]
cpu=["torch", "torchvision"]

[[tool.poetry.source]]
name="pytorch-cuda"
url="https://download.pytorch.org/whl/cu121"
priority="explicit"

[[tool.poetry.source]]
name="pytorch-cpu"
url="https://download.pytorch.org/whl/cpu"
priority="explicit"

You can no basically install with cpu, gpu or no dependency with poetry install -E cpu --with cpu, poetry install -E cuda --with cuda, and poetry install. Adding package-mode=false takes care of only dependency management. Note that this is with the latest poetry version (1.8.3) I have.

AvailableHost · 2024-06-14T22:31:02Z

Do anyone have Simple guide How to update everything (pip install torch, torch2.1.2+cu and whatever) for using SD XL and other model? (after deleting venv folder. When everything fuck's up im just reistall venv)
In my version stable diffusion (amd) every time im try to update SD i have exception
xformers requires torch 2.1.2 but have 1.13......
torchvision requires torch 2.0.0 but have but have 1.13......
...
All i need is 5 line what i need to install in cmd:
pip install torch....
pip install torchvision....
pip install protobuf<4.0...

stf976 · 2024-06-18T17:38:24Z

@shuvamg007 For anyone still looking for a solution, the following hack worked for me on Poetry version 1.8.2:

With Poetry 1.8.3 this does not work. It just always installs the version from the [tool.poetry.dependencies] section (cpu flavor in this case).

@shuvamg007 Adding package-mode=false takes care of only dependency management. Note that this is with the latest poetry version (1.8.3) I have.

This does not work with poetry 1.8.3 either. With the given toml file (using Python 3.11), poetry install simply does not install any packages, regardless of the -E or --with options.

The lock file contains two [[package]] blocks for torchvision, one with the cpu flavor, and the other for cuda. The extras blocks looks like this.

[extras]
cpu = []
cuda = []

I am using Poetry installed from the installer (https://github.com/python-poetry/install.python-poetry.org) on Ubuntu 23.04. Could you please elaborate on what version you are using (OS, installer source)?

lucaspar · 2024-06-18T18:12:35Z

@stf976 there's a slight variation of that pyproject.toml that works, but it's buried by other comments in this thread.

I've created this repo to help others find a setup that works for CPU and GPU installs. See if you find it useful.

shuvamg007 · 2024-06-18T19:10:22Z

I am using Poetry installed from the installer (https://github.com/python-poetry/install.python-poetry.org) on Ubuntu 23.04. Could you please elaborate on what version you are using (OS, installer source)?

Thanks for pointing this out. I just debugged and realized that there needs to be a dependency that needs to force torch to be installed. I'm not sure of this is a bug or not, but I don't have free time to actually debug this (commenting this from work). So as a workaround, I published a package to pypi that imposes this requirement. Please check the following code (hopefully it works):

[tool.poetry]
...
package-mode = false

[tool.poetry.dependencies]
python = "~3.12"

[tool.poetry.group.cpu]
optional=true

[tool.poetry.group.cpu.dependencies]
torch={ version="~2.2", source="pytorch-cpu", markers="extra=='cpu' and extra!='cuda'" }
torchvision={ version="~0.17", source="pytorch-cpu", markers="extra=='cpu' and extra!='cuda'" }
poetorch={ version="0.1.0", extras=["cpu", "vision"] }

[tool.poetry.group.cuda]
optional=true

[tool.poetry.group.cuda.dependencies]
torch={ version="~2.2", source="pytorch-cuda", markers="extra=='cuda' and extra!='cpu'" }
torchvision={ version="~0.17", source="pytorch-cuda", markers="extra=='cuda' and extra!='cpu'" }
poetorch={ version="0.1.0", extras=["cuda", "vision"] }

[tool.poetry.extras]
cuda=[]
cpu=[]

[[tool.poetry.source]]
name="pytorch-cuda"
url="https://download.pytorch.org/whl/cu121"
priority="explicit"

[[tool.poetry.source]]
name="pytorch-cpu"
url="https://download.pytorch.org/whl/cpu"
priority="explicit"

A few things of note:

Installation is as usual. You can install with cpu, gpu or no dependency with poetry install -E cpu --with cpu, poetry install -E cuda --with cuda, and poetry install respectively.
tool.poetry.extras itself doesn't need anything in it. The extras marker needs to exist though. So make sure it stays
The package poetorch has some extras mentioned in it. Depending on which installation you want, add that marker to extras (cpu or cuda). Since not everyone requires torchvision, that was added as an optional extra with the marker vision. Exclude that if you only require torch.
poetorch itself is just an empty project. All it does is add the dependency for torch and torchvision. There are some requirements (python>=3.8, torch==2.*, torchvision==0.*). I'll add the repository to github for reference later and update it here.

Hope this helps

stf976 · 2024-06-20T13:59:26Z

Thanks for pointing this out. I just debugged and realized that there needs to be a dependency that needs to force torch to be installed. I'm not sure of this is a bug or not, but I don't have free time to actually debug this (commenting this from work). So as a workaround, I published a package to pypi that imposes this requirement. Please check the following code (hopefully it works):

With the additional package it works here too (tested with poetry 1.8.3), thanks a lot! It is now possible to switch between torch flavors and install the right packages using the --with cpu and --with cuda parameters.

I also tried it with newer versions of pytorch (torch-2.3.1 and torchvision-0.18.1), and that worked as well.

Unfortunately, poetry export -f requirements.txt --output requirements.txt always outputs the cuda torch packages, regardless of the -E and --with parameters. It does output the right poetorch package though, i.e. with [cpu] or [cuda] options. Using --only cpu or --without cuda doesn't help either. This appears to be a bug in poetry, as it should export only packages from the selected groups as far as I understand.

Overall it seems like a handy solution to the open issue, but it does not work with the current version of poetry in all cases.

ZekunZh · 2024-07-02T16:27:19Z

Inspired by the solution of @shuvamg007, I found another solution that I think is easier to read and understand.

The main idea is to specify only the versions of torch and torchvision in [tool.poetry.dependencies], and switch the source using poetry group and extras.

I tested the pyproject.toml using poetry of version 1.7.1 and version 1.8.3.

To install torch-cpu:

poetry install -E cpu --with cpu --sync

To install torch-gpu:

poetry install -E gpu --with gpu --sync

Hope this can help !

[tool.poetry]
...
package-mode = false

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

[tool.poetry.dependencies]
python = "^3.11"
torch = { version = "^2.2.1", optional = true }
torchvision = { version="^0.17.1", optional = true }

[tool.poetry.group.cpu]
optional = true

[tool.poetry.group.cpu.dependencies]
torch = { version = "^2.2.1", source = "torch_cpu", markers = "extra=='cpu' and extra!='gpu'" }
torchvision = { version="^0.17.1", source = "torch_cpu", markers = "extra=='cpu' and extra!='gpu'" }

[tool.poetry.group.gpu]
optional = true

[tool.poetry.group.gpu.dependencies]
torch = { version = "^2.2.1", source = "torch_cuda", markers = "extra=='gpu' and extra!='cpu'" }
torchvision = { version="^0.17.1", source = "torch_cuda", markers = "extra=='gpu' and extra!='cpu'" }


[tool.poetry.extras]
cpu = ["torch", "torchvision"]
gpu = ["torch", "torchvision"]

[[tool.poetry.source]]
name = "torch_cpu"
url = "https://download.pytorch.org/whl/cpu"
priority = "explicit"

[[tool.poetry.source]]
name = "torch_cuda"
url = "https://download.pytorch.org/whl/cu118"
priority = "explicit"

neviaumi · 2024-07-31T15:42:29Z

I have play around on install PyTorch on my poetry project.

Solution about It only work on 2.2.2

after that it would just show message Unable to find installation candidates for torch (2.3.1)

The configuration i am using

Mac OS 12
Python 3.12
Poetry 1.8

stf976 · 2024-08-01T15:50:52Z

@stf976 there's a slight variation of that pyproject.toml that works, but it's buried by other comments in this thread.

I've created this repo to help others find a setup that works for CPU and GPU installs. See if you find it useful.

Thank you for the setup, I have tried your repo and it works as described. I found the issue I encountered is really related to the lock file.

If a lock file is created with poetry lock before running poetry install --sync -E cpu, the installation always defaults to the cuda packages. As a consequence, export to requirements.txt does not work either, as it always creates a lock file first.

When there is not lock file present, the installation works as described in your repo.

Inspecting the lock file shows both cpu and cuda packages are listed, but it seems there is no way to know for poetry which belongs to the cpu extras and which to cuda, because both packages have the same name and no related metadata.

So in summary, the installation using your config works as described. Combining the config with lock files or exporting to requirements.txt is not possible right now afaics.

brian316 · 2024-08-02T21:18:16Z

this seems to work for me

poetry 1.8.3

[tool.poetry.dependencies]
torch = { version = "2.1.0", source = "pytorch" }

[[tool.poetry.source]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu121"
priority = "explicit"

havardox · 2024-08-18T11:11:07Z

Anyone got a working pyproject.toml for ROCm?

extraymond · 2024-08-23T05:45:45Z

Anyone got a working prproject.toml for ROCm?

After some trial and error I think I know why it sometimes works and sometimes won't when using on rocm.

poetry works, but pytorch didn't have prebuild wheels for every combination of torch/rocm/python.
You can manually check that here: https://download.pytorch.org/whl/torch

For example, only torch 2.4 are build for rocm6.1, the old one are kept in rocm6.0 and rocm5.6.
Let's say you want torch=^2.0 and torchvision installed, which translates to finding torch on the 2.0 series and whatever torchvision that works with torch. I think the resolver will do the following:

find the torch version available in that source
cpu and cuda version is in the same source, they are only differentiate with each other by markers like +cpu +rocm6.1
the resolver find the highest available version with ^2.0 and found one build for cuda
it tried to resolve the torchvision version compatible with torch
and tried to find the rocm variant which aren't available

So when using rocm, I think manually check the version available in the source and directly specify it will works better.

Here's my declaration:

[tool.poetry.dependencies]
torch = { version = "^2.0", source = "torch-rocm" }
torchvision = { version = "^0.16", source = "torch-rocm" }

[[tool.poetry.source]]
name = "torch-rocm"
url = "https://download.pytorch.org/whl/rocm5.6"
priority = "explicit"

be sure to check the python version marker too! if one aren't build for rocm you will face the same issue.

davidgilbertson added area/docs Documentation issues/improvements status/triage This issue needs to be triaged labels Sep 5, 2022

davidgilbertson mentioned this issue Sep 5, 2022

pytorch and poetry #4231

Closed

RonMcKay added a commit to RonMcKay/UQGAN that referenced this issue Sep 30, 2022

Revert to requirements.txt from pyproject.toml

fdf5866

Poetry is not yet ready to handle the different versions of PyTorch and Torchvision. See the related issue: python-poetry/poetry#6409

neersighted mentioned this issue Oct 1, 2022

Backend-dependent PyTorch versions #6671

Closed

2 tasks

dimbleby mentioned this issue May 21, 2024

How to toggle torch cpu and gpu installation using poetry #9430

Closed

david-waterworth mentioned this issue May 30, 2024

Ability to specify a "target" for non-package mode #9456

Open

dimbleby mentioned this issue Jul 1, 2024

Poetry + explicit source can't find installation candidates while pip with --index-url can #9524

Open

ditsuke mentioned this issue Jul 2, 2024

build(python): Package scripts with pep-0517 compliance ggerganov/llama.cpp#5745

Merged

Holt59 mentioned this issue Jul 8, 2024

Poetry lock does not include explicit sources for conditional packages #9537

Open

This was referenced Jul 11, 2024

Support for conflicting dep extras, when they are exclusive of each other. #6419

Open

Pass Install Extras to Markers #9553

Open

dimbleby mentioned this issue Jul 19, 2024

uv add rejects pytorch from its index when on mac astral-sh/uv#5182

Open

asim29 mentioned this issue Sep 4, 2024

Update pyproject.toml file for packaging ssg-research/amulet#33

Merged

Instructions for installing PyTorch #6409

Instructions for installing PyTorch #6409

Comments

davidgilbertson commented Sep 5, 2022 • edited Loading

Issue

Option 1 - wheel URLs for a specific platform

Option 2 - alternate source

dimbleby commented Sep 5, 2022

variolam commented Sep 6, 2022

neersighted commented Sep 6, 2022

Queuecumber commented Sep 6, 2022 • edited Loading

radoering commented Sep 6, 2022

neersighted commented Sep 6, 2022 • edited Loading

Queuecumber commented Sep 6, 2022

Queuecumber commented Sep 6, 2022 • edited Loading

neersighted commented Sep 6, 2022

Queuecumber commented Sep 6, 2022 • edited Loading

ZetiMente commented Sep 24, 2022

neersighted commented Sep 25, 2022

voegtlel commented Oct 12, 2022

neersighted commented Oct 12, 2022

felix-ht commented Oct 18, 2022

neersighted commented Oct 18, 2022

MaKaNu commented Oct 20, 2022

felix-ht commented Oct 20, 2022 • edited Loading

timothyjlaurent commented Nov 3, 2022

brochier commented Nov 10, 2022

neersighted commented Nov 10, 2022

robinbrochier commented Nov 10, 2022

neersighted commented Nov 10, 2022

arthur-st commented Nov 10, 2022

neersighted commented Nov 10, 2022

PyroGenesis commented May 15, 2024

What is possible regarding torch & poetry

slashtechno commented May 15, 2024

QuentinSoubeyranAqemia commented May 16, 2024

david-waterworth commented May 28, 2024

QuentinSoubeyranAqemia commented May 29, 2024 • edited Loading

david-waterworth commented May 30, 2024 • edited Loading

QuentinSoubeyranAqemia commented May 30, 2024

david-waterworth commented May 30, 2024

tfha commented Jun 12, 2024 • edited Loading

shuvamg007 commented Jun 12, 2024 • edited Loading

AvailableHost commented Jun 14, 2024 • edited Loading

stf976 commented Jun 18, 2024

lucaspar commented Jun 18, 2024 • edited Loading

shuvamg007 commented Jun 18, 2024 • edited Loading

stf976 commented Jun 20, 2024 • edited Loading

ZekunZh commented Jul 2, 2024 • edited Loading

neviaumi commented Jul 31, 2024 • edited Loading

stf976 commented Aug 1, 2024 • edited Loading

brian316 commented Aug 2, 2024

havardox commented Aug 18, 2024 • edited Loading

extraymond commented Aug 23, 2024

davidgilbertson commented Sep 5, 2022 •

edited

Loading

Queuecumber commented Sep 6, 2022 •

edited

Loading

neersighted commented Sep 6, 2022 •

edited

Loading

Queuecumber commented Sep 6, 2022 •

edited

Loading

Queuecumber commented Sep 6, 2022 •

edited

Loading

felix-ht commented Oct 20, 2022 •

edited

Loading

QuentinSoubeyranAqemia commented May 29, 2024 •

edited

Loading

david-waterworth commented May 30, 2024 •

edited

Loading

tfha commented Jun 12, 2024 •

edited

Loading

shuvamg007 commented Jun 12, 2024 •

edited

Loading

AvailableHost commented Jun 14, 2024 •

edited

Loading

lucaspar commented Jun 18, 2024 •

edited

Loading

shuvamg007 commented Jun 18, 2024 •

edited

Loading

stf976 commented Jun 20, 2024 •

edited

Loading

ZekunZh commented Jul 2, 2024 •

edited

Loading

neviaumi commented Jul 31, 2024 •

edited

Loading

stf976 commented Aug 1, 2024 •

edited

Loading

havardox commented Aug 18, 2024 •

edited

Loading