How to install python package directly from GitHub?

[asmith26]

What's the equivalent to pip install git+https://github.com/huggingface/parler-tts.git?

[asmith26]

This seems to work/be the correct format: pixi add --pypi "parler-tts @ git+https://github.com/huggingface/parler-tts.git"

[Shobhit0109]

How can we install globally from github?

[matthewfeickert]

@Shobhit0109 at the moment pixi global doesn't have pipx like functionality and it isn't on the roadmap given #565 (comment). So to achieve what you want you'd do

# pixi global install git  # if you somehow don't have Git
$ pixi global install pipx
$ pipx install 'git+https://github.com/huggingface/parler-tts.git'

However, if you do that you'll get the notice

No apps associated with package parler_tts. Try again with '--include-deps' to include apps of dependent packages, which are listed above. If you are attempting to install a library, pipx should not be used.
Consider using pip or a similar tool instead.

so apparently parler-tts doesn't provide any CLI API. Given this, I'm not sure what your use case is of wanting to try to install it globally.

[PiotrCzapla]

Hi guys this doesn't seems to work anymore:

pixi add --pypi "parler-tts @ git+https://github.com/huggingface/parler-tts.git"
  × Expected package name starting with an alphanumeric character, found `@`
  │ @
  │ ^

[matthewfeickert]

@PiotrCzapla That works for me on

$ pixi --version
pixi 0.33.0

$ pixi init
✔ Created /tmp/pixi-debug/pixi.toml
$ pixi add 'python<3.13'
✔ Added python<3.13
$ pixi add --pypi "parler-tts @ git+https://github.com/huggingface/parler-tts.git"
  × failed to solve the pypi requirements of 'default' 'linux-64'
  ╰─▶ Build backend failed to determine metadata through `prepare_metadata_for_build_wheel` with exit status: 1
      --- stdout:
      
      --- stderr:
      Traceback (most recent call last):
        File "<string>", line 14, in <module>
        File "/home/feickert/.cache/rattler/cache/uv-cache/builds-v0/.tmpKO6HKs/lib/python3.12/site-packages/setuptools/build_meta.py", line 373, in prepare_metadata_for_build_wheel
          self.run_setup()
        File "/home/feickert/.cache/rattler/cache/uv-cache/builds-v0/.tmpKO6HKs/lib/python3.12/site-packages/setuptools/build_meta.py", line 516, in run_setup
          super().run_setup(setup_script=setup_script)
        File "/home/feickert/.cache/rattler/cache/uv-cache/builds-v0/.tmpKO6HKs/lib/python3.12/site-packages/setuptools/build_meta.py", line 318, in run_setup
          exec(code, locals())
        File "<string>", line 354, in <module>
        File "<string>", line 87, in get_ext_modules
      ModuleNotFoundError: No module named 'numpy.distutils'
      ---

(where this is obviously failing as numpy.distutils has been deprecated for some time, but that's not a pixi issue.)

[SheldonWBM]

I am experimenting with Pixi and seeing which projects a Pixi package manager can be applied to.

pixi add --PyPI "label-studio-ml @ git+https://github.com/HumanSignal/label-studio-ml-backend.git"
  x failed to solve the pypi requirements of 'default' 'linux-64'
  |-> Failed to download and build: `label-studio-sdk @ git+https://github.com/HumanSignal/label-studio-sdk.git`
  |-> Failed to extract static metadata from `pyproject.toml`
  |-> `pyproject.toml` is using the `[project]` table, but the required `project.name` field is not set.
  `-> TOML parse error at line 30, column 2
         |
      30 | [project.urls]
         |  ^^^^^^^
      missing field `name`

When I move the dependency from [pypi-dependencies] to [dependencies] I have this error

label-studio-ml = { git = "https://github.com/HumanSignal/label-studio-ml-backend.git" }
    ---- source dependencies are not allowed yet

label-studio-ml is out-of-date on PyPi and does not exist in Conda. So, for this project. Being able to install from GitHub or a zip file is essential.

I am checking out Pixi (for fun), as a guest speaker mentioned it in a Posit Data Science Hangout.

[SheldonWBM]

hist, numpy have a [project] field in the pyproject.toml for example:

[project]
name = "numpy"
version = "2.2.0.dev0"
# TODO: add `license-files` once PEP 639 is accepted (see meson-python#88)
license = {file = "LICENSE.txt"}

description = "Fundamental package for array computing in Python"
authors = [{name = "Travis E. Oliphant et al."}]
maintainers = [
    {name = "NumPy Developers", email="[email protected]"},
]
requires-python = ">=3.10"

The label-studio-sdk repository does not yet have a [project] field, it has a [tool.poetry] field with similar info.

[matthewfeickert]

The label-studio-sdk repository does not yet have a [project] field, it has a [tool.poetry] field with similar info.

@SheldonWBM not quite. It uses [project.urls], as the error message shows, which is invalid without the rest of the project table. Poetry doesn't adhere to the standards and does its own thing, which causes problems like this.

[SheldonWBM]

Update: I was able to work around it.

[tasks]
install-label-studio-ml = "pip install git+https://github.com/HumanSignal/label-studio-ml-backend.git"
install-label-studio-sdk = "pip install git+https://github.com/HumanSignal/label-studio-sdk.git"

The label-studio-ml the package is now detected. I can pixi run the ml backend.

It could be a more elegant solution, but it works in the docker container.

[matthewfeickert]

Glad that works well enough for you now. Though note that these dependencies aren't getting recorded in the pypi-dependencies table in pixi.toml or in the pixi.lock, and this makes it harder to take advantage of the nice patterns for deployment of pixi environments to Linux containers.

[SheldonWBM]

I am investigating using Pixi for that reason. I want to clean-up my dockerfile and have all packages installed through Pixi.
Due to dependency version conflicts, the packages are messy with a specific and fragile install order. I am hoping I can use the pixi.toml to at least decrease the mess. The lable-studio packages are not in the pixi.lock but they are installed in the environment (not globally). I want one package manager to rule them all.