A technical paper titled “Lean Attention: Hardware-Aware Scalable Attention Mechanism for the Decode-Phase of Transformers” was published by researchers at Microsoft. “Transformer-based models have ...