Accelerated Inference with Optimum and Transformers Pipelines