Can't install python=3.12 and torchvision=0.19.1 with pixi, but works with mamba

[asmith26]

Checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pixi, using pixi --version.

Reproducible example

Unfortunately this doesn't work:

[project]
authors = [{name = "name", email = "email@email.com"}]
dependencies = []
description = "Add a short description here"
name = "test_env"
requires-python = ">= 3.12"
version = "0.1.0"

[tool.pixi.project]
channels = ["pytorch", "conda-forge"]
platforms = ["linux-64"]

[tool.pixi.dependencies]
python = ">=3.12"
torchvision = ">=0.19.1"

running pixi install yields:

  × failed to solve the conda requirements of 'default' 'linux-64'
  ╰─▶ Cannot solve the request because of: torchvision >=0.19.1 cannot be installed
      because there are no viable options:
      ├─ torchvision 0.19.1 | 0.19.1 | 0.19.1 | 0.19.1 | 0.19.1 would require
      │  └─ pytorch-cuda 12.1.*, which cannot be installed because there are no
      viable options:
      │     └─ pytorch-cuda 12.1 would require
      │        └─ libcublas >=12.1.0.26,<12.1.3.1, for which no candidates were
      found.
      └─ torchvision 0.19.1 | 0.19.1 | 0.19.1 | 0.19.1 | 0.19.1 would require
         └─ pytorch-cuda 11.8.*, which cannot be installed because there are no
      viable options:
            └─ pytorch-cuda 11.8 | 11.8 | 11.8 would require
               └─ cuda 11.8.*, for which no candidates were found.
      python >=3.12 cannot be installed because there are no viable options:
      └─ python 3.13.0rc1 | 3.13.0rc1 | 3.13.0rc1 | 3.13.0rc1 | 3.13.0rc1 |
      3.13.0rc1 would require
         └─ _python_rc *, for which no candidates were found.
      The following packages are incompatible
      ├─ torchvision >=0.19.1 can be installed with any of the following options:
      │  ├─ torchvision 0.19.1 | 0.19.1 would require
      │  │  └─ pillow >=5.3.0,!=8.3.*, which can be installed with any of the
       following options:
      │  │     ├─ pillow 9.2.0 | 9.2.0 | 9.4.0 would require
      │  │     │  └─ python_abi 3.11.* *_cp311, which can be installed with any of
      the following options:
      │  │     │     └─ python_abi 3.11
      │  │     ├─ pillow 8.4.0 | 9.0.0 | 9.0.1 | 9.0.1 | 9.0.1 | 9.1.0 | 9.1.0 |
      9.1.0 | 9.1.1 | 9.1.1 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 | 9.4.0 would
      require
      │  │     │  └─ python_abi 3.10.* *_cp310, which can be installed with any of
      the following options:
      │  │     │     └─ python_abi 3.10
      │  │     ├─ pillow 7.2.0 | 8.0.0 | 8.0.1 | 8.0.1 | 8.1.0 | 8.1.0 | 8.1.0 |
      8.1.1 | 8.1.2 | 8.1.2 | 8.2.0 | 8.4.0 | 9.0.0 | 9.0.1 | 9.0.1 | 9.0.1 | 9.1.0
      | 9.1.0 | 9.1.0 | 9.1.1 | 9.1.1 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 |
      9.4.0 would require
      │  │     │  └─ python_abi 3.9.* *_cp39, which can be installed with any of the
      following options:
      │  │     │     └─ python_abi 3.9 | 3.9
      │  │     ├─ pillow 6.2.2 | 7.0.0 | 7.1.1 | 7.1.2 | 7.2.0 | 7.2.0 | 7.2.0 |
      8.0.0 | 8.0.1 | 8.0.1 | 8.1.0 | 8.1.0 | 8.1.0 | 8.1.1 | 8.1.2 | 8.1.2 | 8.2.0
      | 8.4.0 | 9.0.0 | 9.0.1 | 9.0.1 | 9.0.1 | 9.1.0 | 9.1.0 | 9.1.0 | 9.1.1 |
      9.1.1 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 | 9.2.0 | 9.4.0 would require
      │  │     │  └─ python_abi 3.8.* *_cp38, which can be installed with any of the
      following options:
      │  │     │     └─ python_abi 3.8 | 3.8 | 3.8
      │  │     ├─ pillow 6.2.2 | 7.0.0 | 7.1.1 | 7.1.2 | 7.2.0 | 7.2.0 | 7.2.0 |
      8.0.0 | 8.0.1 | 8.0.1 | 8.1.0 | 8.1.0 | 8.1.0 | 8.1.1 | 8.1.2 | 8.1.2 | 8.2.0
      | 8.4.0 | 9.0.0 | 9.0.1 | 9.0.1 | 9.0.1 | 9.1.0 | 9.1.0 | 9.1.0 | 9.1.1 |
      9.1.1 | 9.2.0 | 9.2.0 | 9.2.0 would require
      │  │     │  └─ python_abi 3.7.* *_cp37m, which can be installed with any of
      the following options:
      │  │     │     └─ python_abi 3.7 | 3.7 | 3.7
      │  │     ├─ pillow 6.2.2 | 7.0.0 | 7.1.1 | 7.1.2 | 7.2.0 | 7.2.0 | 7.2.0 |
      8.0.0 | 8.0.1 | 8.0.1 | 8.1.0 | 8.1.0 | 8.1.0 | 8.1.1 | 8.1.2 | 8.1.2 | 8.2.0
      would require
      │  │     │  └─ python_abi 3.6.* *_cp36m, which can be installed with any of
      the following options:
      │  │     │     └─ python_abi 3.6 | 3.6 | 3.6
      │  │     └─ pillow 6.2.1 | 6.2.1 | 7.0.0 would require
      │  │        └─ python >=3.8,<3.9.0a0, which can be installed with any of the
      following options:
      │  │           └─ python 3.8.14 | 3.8.14 | 3.8.15 | 3.8.15 | 3.8.15 | 3.8.16 |
      3.8.16 | 3.8.17 | 3.8.18 | 3.8.18 | 3.8.19
      │  ├─ torchvision 0.19.1 | 0.19.1 would require
      │  │  └─ python >=3.10,<3.11.0a0, which can be installed with any of the
      following options:
      │  │     └─ python 3.10.9 | 3.10.11
      │  ├─ torchvision 0.19.1 | 0.19.1 would require
      │  │  └─ python >=3.9,<3.10.0a0, which can be installed with any of the
      following options:
      │  │     └─ python 3.9.14 | 3.9.14 | 3.9.15 | 3.9.16 | 3.9.16 | 3.9.17 |
      3.9.17 | 3.9.18 | 3.9.18 | 3.9.18 | 3.9.18 | 3.9.19
      │  └─ torchvision 0.19.1 | 0.19.1 would require
      │     └─ python >=3.8,<3.9.0a0, which can be installed with any of the
      following options:
      └─ python >=3.12 cannot be installed because there are no viable options:
         ├─ python 3.12.5, which conflicts with the versions reported above.
         └─ python 3.12.0 | 3.12.1 | 3.12.1 | 3.12.2 | 3.12.3 | 3.12.4, which
      conflicts with the versions reported above.

Issue description

The compatible package does indeed appear to exist: https://anaconda.org/pytorch/torchvision/files?version=0.19.1

Expected behavior

This works (e.g. I can successfully import torchvision)

mamba create -n test_env python=3.12 torchvision=0.19.1 -c pytorch -c conda-forge

[tdejager]

I can't figure out why it cannot install just yet (@ruben-arts) has looked more into torch issues, he might have some insight. But at least this is working for me:

[project]
dependencies = []
description = "Add a short description here"
name = "test_env"
version = "0.1.0"
requires-python = ">= 3.12"

[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64"]

[tool.pixi.dependencies]
torchvision = ">=0.19.0"

This does depend on a patch version less than the one that you proposed but is making use of the conda-forge channel exclusively.

[ruben-arts]

There are two ways to fix this:

• Verbose but correct way, set the channels per dependency that is not coming from the first channel:

#...
[tool.pixi.project]
# Switch channels to make conda-forge the highest priority
channels = ["conda-forge","pytorch"]
platforms = ["linux-64"]

[tool.pixi.dependencies]
python = ">=3.12"
# Define all packages you want to fetch from 'pytorch' with a specific channel.
# You could leave the once that only exist on 'pytorch' channel out but this is the safest way. 
pytorch = {channel = "pytorch", version = "*"}
pytorch-cuda = {channel = "pytorch", version = "*"}
pytorch-mutex = {channel = "pytorch", version = "*"}
torchtriton = {channel = "pytorch", version = "*"}
torchvision = {channel = "pytorch", version = ">=0.19.0"}

• Fast but possibly dangerous way, disable channel-priority.

#...
[tool.pixi.project]
channels = ["conda-forge","pytorch"]
# Let pixi use any mix of packages from all channels and just give any solution. 
channel-priority = "disabled"
platforms = ["linux-64"]

[tool.pixi.dependencies]
python = ">=3.12"
torchvision = ">=0.19.0"

ps. don't forget to set the cuda system-requirements:

[tool.pixi.system-requirements]
cuda = "12.1"

[asmith26]

Thanks very much for your help @tdejager @ruben-arts

This works for me:

[project]
authors = [{name = "name", email = "email@email.com"}]
dependencies = []
description = "Add a short description here"
name = "test_env"
requires-python = ">= 3.12"
version = "0.1.0"

[tool.pixi.project]
channels = ["conda-forge","pytorch"]
platforms = ["linux-64"]

[tool.pixi.dependencies]
python = ">=3.12"
pytorch = {channel = "pytorch", version = "*"}
torchvision = {channel = "pytorch", version = ">=0.19.1"}