Training loss - how to manage

Hi, I have something around 200 + intents and average about 100+ examples per intent. I am using Rasa NLU 0.13.0.a2 on Linux. I am using TF embedding pipeline and for the first time, I am facing an issue where training loss seems to be high (~ 0.25). Though I get accuracy pretty close to 1 (at around 10-12 epochs), losses were high at around 0.28. I trained upto 1000 epochs and still losses stay around 0.248., where as training accuracy also dropped to 0.940 as I moved towards 1000 epochs. What could be wrong? Any pointers from your experience?

why do you think this loss is too high, with what do you compare it? If accuracy is high enough, I don’t see a problem