pip install tensorflow could not find a version

[alexcreature]

System information

• macOS Catalina version 10.15.3
• pip install tensorflow
• GPU model and memory: Intel Iris Plus Graphics 645 1536 MB

The problem
Both pip install tensorflow and pip install tf-nightly lead to the following error messages:

ERROR: Could not find a version that satisfies the requirement tensorflow (from versions: none)

ERROR: No matching distribution found for tensorflow

[ymodak]

Did you try updating your pip version?

[amahendrakar]

@alexcreature.

As per the documentation TensorFlow 2 packages require a pip version >19.0.

Could you please try the following commands and let us know if it works.
python3 -m pip install --upgrade pip or pip install --upgrade pip and then try

pip install tensorflow

Thanks!

[dA505819]

@alexcreature Try updating pip version by:

pip install --upgrade pip

and the you can again install tensorflow by:

pip install tensorflow

or you can specify version which you want to install.

This seems to be the same as an issue I had previously with both Windows 10 and Ubuntu 18 and Ubuntu 16, no matter the operating system.
The previous issue ( #37205 ) was closed because I proposed a WORK AROUND of downloading the .whl and installing Tensorflow through the downloaded wheel, which worked just fine for me.

Something is broken in the pypi pipeline of installation and needs to be taken care of...

[mihaimaruseac]

Yes, the issue is that systems use old pip / setuptools by default and these don't know how to install packages with manylinux2010 tags. That's why you need to upload pip and setuptools first (and always install in a virtualenv, not globally)

[amahendrakar]

@alexcreature,
Any updates regarding this issue? Thanks!

Yes, the issue is that systems use old pip / setuptools by default and these don't know how to install packages with manylinux2010 tags. That's why you need to upload pip and setuptools first (and always install in a virtualenv, not globally)

Okay so here are ALL of the things I have tried in my attempts to simply install tensorflow via pip install:

pip 20.0.2
Python 3.7.6
Win10, globally
Ubuntu 18.04, globally (source built python 3.7.6 and alias setup to avoid breaking apt)
Ubuntu 16.xx, globally, same as above
Win10 pipenv environment (most recent version of pipenv)
Ubuntu 16, 18 pipenv environment (most recent version of pipenv)

I have tried and tested everything I can possibly come up with to make things different but nothing seems to install Tensorflow 2.1 for Python 3.7 for me, in any way shape or form. I have tried pip install tensorflow I have tried pip install tensorflow==2.1 and the absolute only thing I can do is install by downloading the wheel and installing that wheel directly.

Is there something we can do with pip or python source to get this back to working properly? I don't have ANY issues with any other packages I use across the board, except for Black, but even then sometimes that will install without a hitch and other times I get the same error, though it's been a long time since I tried and that is not relevant to Tensorflow...

Edit: I cannot find any PEP which shows a breaking change in manylinux tags and pip installs... The most relevant is 571 which states that any pip that is capable of installing manylinux2010 tags will be backwards compatible, and 600 which shows a new naming convention for manylinux tags, but also states that old tags should still work, if I'm reading correctly.
PEP 599 also adds the manylinux2014 tag, however in true python fashion, retains support for manylinux1 and manylinux2010 tags, so this definately should not be the issue, however, I am no source developer and am simply relaying the information I have found.....

These are the most relevant PEP's I have found regarding manylinux wheel tags.

[mihaimaruseac]

I think the best way forward now would be to open an issue on PyPI https://github.com/pypa/pip

If you do, please link against this issue and/or tag me too to follow up

[amahendrakar]

@alexcreature,
Any updates regarding this issue? Thanks!

Automatically closing due to lack of recent activity. Please update the issue when new information becomes available, and we will reopen the issue. Thanks!