The structure of the contents of a requirements.txt file is defined as follows:
[[--option]...]
<requirement specifier> [; markers] [[--option]...]
<archive url/path>
[-e] <local project path>
[-e] <vcs project url>
The <requirement specifier>
defines the package and an optional version.
SomeProject
SomeProject == 1.3
SomeProject >=1.2,<2.0
SomeProject[foo, bar]
SomeProject~=1.4.2
The --option
(such as the -f
/--find-links
) is the same as the pip install options you would use if you were doing pip install
from the command line.
The following options are supported:
- -i, --index-url
- --extra-index-url
- --no-index
- -c, --constraint
- -r, --requirement
- -e, --editable
- -f, --find-links
- --no-binary
- --only-binary
- --require-hashes
- --pre
- --trusted-host
So, for your install commands, the requirements.txt would look like this:
# Torch
--find-links https://download.pytorch.org/whl/torch_stable.html
torch==1.5.0+cu101
torchvision==0.6.0+cu101
# Detectron
--find-links https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/index.html
detectron2
Make sure to verify that the links are correctly used:
$ pip install -r requirements.txt
Looking in links: https://download.pytorch.org/whl/torch_stable.html, https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/index.html
Collecting torch==1.5.0+cu101 (from -r requirements.txt (line 3))
Using cached https://download.pytorch.org/whl/cu101/torch-1.5.0%2Bcu101-cp38-cp38-linux_x86_64.whl
Collecting torchvision==0.6.0+cu101 (from -r requirements.txt (line 4))
Using cached https://download.pytorch.org/whl/cu101/torchvision-0.6.0%2Bcu101-cp38-cp38-linux_x86_64.whl
Collecting detectron2 (from -r requirements.txt (line 8))
Using cached https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/detectron2-0.1.2%2Bcu101-cp38-cp38-linux_x86_64.whl
...
As a side note, you originally said "(not github)" in your title. The default source of packages installed using pip
is hosted on PyPi: https://files.pythonhosted.org/. You can see the actual links when going to the Download Files section of a package in PyPi (example for Torch).