Text this: Optimizing the Learnable RoPE Theta Parameter in Transformers