Feature: reset/override value head weights while keeping actor intact (warm-start / continual learning)

Hello,

I’d like to reset or modify only the value function (critic) head of a trained policy while keeping the actor weights unchanged, so I can warm-start a new training phase that adapts the critic to distribution shifts (new reward scales, curriculum stage, etc.) without perturbing the policy. Is there a supported way to partially re-initialize / load only the critic parameters? If yes, could you please provide guidance/pointers for that?

Thank you!