What is the pixi equilvalent to pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu?

[asmith26]

Hi, I'm trying to install a CPU version of (py)torch using the following command, but I can't seem to get pixi to install torch from https://download.pytorch.org/whl/cpu.

What is the pixi equilvalent to the following (I'm using a pyproject.toml file)?

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

Many thanks for any help!

[asmith26]

I've got it working:

Option 1 (easier)

Following https://pixi.sh/v0.22.0/reference/configuration/#the-pypi-options-table, add to pyproject.toml:

[tool.pixi.pypi-options]
index-url = "https://pypi.org/simple"
extra-index-urls = ["https://download.pytorch.org/whl/cpu"]

Option 2 (more work)

Following https://pixi.sh/v0.22.0/reference/configuration/#pypi-dependencies, pass in the appropriate URL (within pyproject.toml, e.g.

torch = { url = "https://download.pytorch.org/whl/cpu/torch-2.4.0%2Bcpu-cp312-cp312-linux_x86_64.whl#sha256=78dbf5f2789933a7ea2dabeead4daa44679b1e0d8eb35ddb7071c8ab7b181eb3" }

---

Thanks again for an amazing lib!

[m0n203]

I am using this way:

pixi.toml

[pypi-dependencies]
torch = { version = "*", index = "https://download.pytorch.org/whl/cpu" }

but it's no way to do this w/ pixi add --pypi, have to edit toml manually.

[Norlandz]

Pytorch Installation - Pixi by prefix.dev
https://pixi.sh/dev/python/pytorch/

I just follow the official docs :

[system-requirements]
cuda = "12.0"

[dependencies]
# pytorch-gpu = ">=2.5.1,<2.8"
pytorch-gpu = "*"
cuda-version = "12.6.*"

Then pixi i
Then pixi add torchvision

More:

[workspace]
channels = ["conda-forge", "pytorch"]
name = "build-llm-demo"
platforms = ["win-64"]
version = "0.1.0"

[system-requirements]
cuda = "12.0"

[dependencies]
python = "3.11.*"
pandas = "<=2.1.4"
pydantic-settings = ">=2.9.1,<3"
ruff = ">=0.12.0,<0.13"
pyright = ">=1.1.402,<2"
pytest = ">=8.4.1,<9"
# pytorch-gpu = ">=2.5.1,<2.8"
pytorch-gpu = "*"
cuda-version = "12.6.*"
ipykernel = ">=6.29.5,<7"
pixi-pycharm = ">=0.0.8,<0.0.9"
ipywidgets = ">=8.1.7,<9"
transformers = ">=4.13,<4.38"
huggingface_hub = ">=0.33.1,<0.34"
matplotlib = ">=3.10.3,<4"
itables = ">=2.4.2,<3"
tiktoken = ">=0.9.0,<0.10"
jupyterlab = ">=4.4.4,<5"
pypandoc = ">=1.15,<2"
pandoc = ">=3.7.0.2,<4"
jupytext = ">=1.17.2,<2"
torchvision = ">=0.22.0,<0.23"


[pypi-dependencies]
fastapi = { version = ">=0.115.13, <0.116", extras = ["standard"] }
uvicorn = { version = ">=0.34.3, <0.35", extras = ["standard"] }
gunicorn = ">=23.0.0, <24"
build-llm-learn = { path = ".", editable = true }
tensorflow = ">=2.18.0, <3"
bpemb = ">=0.3.6,<0.4"

[Norlandz]

There is another way from pypi:

Pytorch Installation - Pixi by prefix.dev
https://pixi.sh/dev/python/pytorch/#pytorch-index

[pypi-dependencies]
torch = { version = ">=2.5.1", index = "https://download.pytorch.org/whl/cu124" }
torchvision = { version = ">=0.20.1", index = "https://download.pytorch.org/whl/cu124" }

(doesnt seem like need cuda as a package or something..)

Full code:

[workspace]
authors = ["norlz <[email protected]>"]
channels = ["conda-forge"]
name = "iopaint-exe"
platforms = ["win-64"]
version = "0.1.0"

[dependencies]
python = "3.10.*"

[pypi-dependencies]
iopaint = "==1.5.3"
torch = { version = ">=2.5.1", index = "https://download.pytorch.org/whl/cu124" }
torchvision = { version = ">=0.20.1", index = "https://download.pytorch.org/whl/cu124" }

[Norlandz]

Update:

In some setup

[system-requirements]
cuda = "12.0"

Seems crucial.

Otherwise I get things like

D:\usp\uhpsj\study\build_llm_wsp\build_llm>pixi i
Error:   × failed to solve requirements of environment 'default' for platform 'win-64'                    
  ├─▶ failed to solve the environment                                                                     
  ╰─▶ Cannot solve the request because of: pytorch-gpu * cannot be installed because there are no viable  
      options:
      ├─ pytorch-gpu 2.7.1 would require
      │  └─ pytorch ==2.7.1 cuda*_mkl*304, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.7.1 would require
      │  └─ pytorch ==2.7.1 cuda*_mkl*303, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.7.1 would require
      │  └─ pytorch ==2.7.1 cuda*_mkl*302, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.7.1 would require
      │  └─ pytorch ==2.7.1 cuda*_mkl*301, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.7.1 would require
      │  └─ pytorch ==2.7.1 cuda*_mkl*300, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 | 2.7.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.7.0 would require
      │  └─ pytorch ==2.7.0 cuda*_mkl*300, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.7.0 | 2.7.0 | 2.7.0 | 2.7.0 | 2.7.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.6.0 would require
      │  └─ pytorch ==2.6.0 cuda*_mkl*304, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.6.0 would require
      │  └─ pytorch ==2.6.0 cuda*_mkl*303, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.6.0 would require
      │  └─ pytorch ==2.6.0 cuda*_mkl*302, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.6.0 would require
      │  └─ pytorch ==2.6.0 cuda*_mkl*301, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.6.0 would require
      │  └─ pytorch ==2.6.0 cuda*_mkl*300, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 | 2.6.0 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*317, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*316, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*315, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*314, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*313, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*312, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*311, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      ├─ pytorch-gpu 2.5.1 would require
      │  └─ pytorch ==2.5.1 cuda*_mkl*310, which cannot be installed because there are no viable
      options:
      │     └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
      │        └─ __cuda *, for which no candidates were found.
      └─ pytorch-gpu 2.5.1 would require
         └─ pytorch ==2.5.1 cuda*_mkl*309, which cannot be installed because there are no viable
      options:
            └─ pytorch 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 | 2.5.1 would require
               └─ __cuda *, for which no candidates were found.