Slow inference while performing translation

I’m using the mbart model to perform the translation of 1000 sentences and I see that the inference is very slow. Each sentence takes about a minute to infer. How do I speed up the inference ( I do have access to GPU).

Any help appreciated,