Hi @ExtremeViscent Thanks for your contribution! Here are some suggestions for improvement:
Organize files, there are many files in the root directory.
Polish README: What is the expectation training result for this example, and did you verify its correctness (please provide the satisfactory log)?
Currently, this example seems only support 2d tensor parallel. Do you consider data/other tensor/pipeline parallel, ZeRO, etc.? Could you please try to support them, provide the configs, and update expected results in README? Thanks.
A Colossal-AI implementation of Large Scale GAN Training for High Fidelity Natural Image Synthesis. Data and multiple tensor parallel mode supported.