What is a good value for loss?

Hi, I have a text dataset. This dataset includes sentences and scores (one score for each sentence). I want to learn these scores by RNN network. I have used gru (Gated Recurrent Network) with one layer. I have used MSE loss and SGD optimizer. The scores of the sentences are at range(0,0.2) and I get loss=50 after 100 epochs with learning rate=0.0005. (at first, my learning rate was 0.001 and I received huge losses). I wanna know how can I know that a loss value is good or bad? what is an appropriate loss value for my work? and if it is bad, how can I improve my network to get smaller loss values?