I am trying to use this custom optimizer: GitHub - ItzikMalkiel/MTAdam: MTAdam: Automatic Balancing of Multiple Training Loss Terms
However I can’t seem to figure out how to integrate it with lighting. I tried using Manual Optimization, but could not get it to work. The challenges are:
- It has a custom step function that requires the loss be passed in
- It calls backwards inside step
Any idea how I could use this with lightning? Or am I better off in this case sticking to regular pytorch?