search

Optimization-AI / LibAUC

LibAUC: A Deep Learning Library for X-Risk Optimization
https://libauc.org/
MIT License
273 stars 37 forks source link

LibAUC code #13

Closed aashiqmuhamed closed 1 year ago

aashiqmuhamed commented 2 years ago

I was wondering whether you could also commit the core library code to this repository? I just see examples and scripts, and no code for the actual library.

Are there plans to integrate the optimizers with other large scale training frameworks such as DeepSpeed? Have you tested the optimizers on large models? (say >500M parameters)

jacobdang commented 2 years ago

It appears that the actual library and code are not yet released. Running the examples according to the provided instruction only gives an error message: AttributeError: module 'libauc' has no attribute 'losses'. Wondering when will the library be released?

yzhuoning commented 2 years ago

@jacobdang The source file is actually available here. You can simply download it and use pip to install. After the installation, the source code is located on your local python directory. We are currently working on a major release and expect to push this update sometime next month.

Regarding your code, it is hard to diagnose why you ran into this error without further information provided (e.g., which version, tutorials, etc.).

lengstrom commented 2 years ago

I think what @jacobdang means here is that the source files for the actual library are not hosted on GitHub -- the GitHub repo and the releases page only contain code for the tutorials/examples not the python module.

tstuessi commented 2 years ago

Seconded, I would also prefer to have the source files for the library on Github - I've been using this library quite a bit and it would be useful to be able to search the source files in a browser.

marcmk6 commented 1 year ago

It would be nice to have source code. I'm currently struggling running one of the examples as it's extremely slow for me to download a dataset. If the source code is available, there can be flexible solutions.

In addition, as mentioned above, transparency of the code will definitely improve the usefulness of the library as it can be combined with various frameworks to quickly conduct experiments with the algorithms developed by your group, in which situation everyone wins I guess.