Closed halr9000 closed 3 months ago
The training code may be open sourced later and the environment configuration will be updated. I closed that PR first.
The training code may be open sourced later and the environment configuration will be updated. I closed that PR first.
Ok, cool, just know that the install steps do not work on some systems as written when tested from a clean Windows install. Doing a conda install without creating an environment and activating it will install to the base environment. At least Conda is good about managing conflicts, so unlikely to cause too many issues, but it's not a best practice. And the inclusion of a requirements.txt file is going to mess up some people. For example in vscode, when you open a Python repo, it scans for that file, and if present, will suggest it be used to create a venv, which will 1/ install things which aren't needed, take long, and might cause some complications (scikit is a great example), and 2/ fail to install the right pytorch w/cuda on an Nvidia system, as you are having people do that via conda. Which would only work -- if the user does not create a conda or venv, because torch will get installed AGAIN, the wrong way, and that incorrect torch will be used on purpose, because that's what venvs are for -- to isolate environments.
This PR does the following:
I've only tested it on Windows, but see no reason it won't work on Linux.