jnp.sign(mean_rms) is always 1

[OhadRubin]

In the list of features for nn_adam, you are using this feature, but if I understand correctly, this feature is always 1.

learned_optimization/learned_optimization/learned_optimizers/nn_adam.py

Line 305 in 242e218

inputs["mean_sign"] = jnp.sign(mean_rms)