xLSTM 7B: A Recurrent LLM for Fast and Efficient Inference