think it is accounted for by this scale the kimi team introduced, so just default to adam lr

Author: lucidrains

adam_atan2_pytorch/muon_adam_atan2.py

@@ -59,7 +59,7 @@ def __init__(
         muon_params,
         params,
         lr = 1e-4,
-        muon_lr = 1e-3,
+        muon_lr = None,
         betas: tuple[float, float] = (0.9, 0.99),
         weight_decay = 0.,
         regen_reg_rate = 0.,

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "adam-atan2-pytorch"
-version = "0.2.3"
+version = "0.2.4"
 description = "Adam-atan2 for Pytorch"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }