I see "Text Encoder Learning Rate" is now an option. Does this also follow the same learning rate schedule (polynomial, cosine, etc)

[morphinapg]

I use a polynomial learning rate schedule with a power of 6 with a lot of steps. I've always wanted the text encoder learning rate to be half what the standard learning rate is. If I set TE Learning rate to half the value I set for the regular learning rate, is this what's happening?

Unfortunately, neither the UI or command line show the current text encoder learning rate.

[morphinapg]

I tried looking through the code. I don't write python myself, so I could easily have missed something that may be more obvious to a python dev, but I looked around, trying to figure out what that text encoder learning rate setting is doing, ultimately looking around for basically any mention of "tenc" in the code, and anything that makes use of functions that make use of it. It seems like it doesn't look like the formulas that are calculated on the non-constant learning rate schedules are doing anything with the tenc_lr, so my guess is that parameter would either be ignored, using the standard lr instead, or it may remain constant, which would be even worse. It seems like it's used with the new adaptive optimizers, and with Lora, but that might be it. Perhaps someone better with Python can look into it to confirm for me.