This last weekend, I attempted to make an old python ML code base working, and it was very frustrating.<p>The code is from 2018, and it is maintained, but using python3.8+torch1.4. However, one of its dependency is not maintained anymore and it doesn't work with newer python.<p>I ran into so many other issues even with conda. Things could get worse if you want to merge two ML models using different torch and python versions into one workflow.<p>This is broken.