According to the qubit report, scholars from Microsoft Asia Research Institute (MSRA) proposed a new large model architecture Retentive Network (RetNet) in the paper "Retentive Network: A Successor to Transformer for Large Language Models", which is regarded as the field of large models Transformer's successor. Experimental data shows that on language modeling tasks: RetNet can achieve perplexity comparable to Transformer, reasoning speed is 8.4 times, memory usage is reduced by 70%, and it has good scalability. And when the model size is larger than a certain scale, RetNet will perform better than Transformer.