Retentive Network: A Successor to Transformer for Large Language Models

TL;DR

Retentive Network (RetNet) introduces a novel architecture for large language models that combines parallel training, low-cost inference, and efficient long-sequence modeling, serving as a promising successor to Transformers.

Contribution

RetNet presents a new sequence modeling mechanism with three computation paradigms, enabling efficient training and inference while maintaining strong performance.

Findings

RetNet achieves favorable scaling in language modeling.

It enables low-cost, high-throughput inference.

The architecture supports efficient long-sequence processing.

Abstract

In this work, we propose Retentive Network (RetNet) as a foundation architecture for large language models, simultaneously achieving training parallelism, low-cost inference, and good performance. We theoretically derive the connection between recurrence and attention. Then we propose the retention mechanism for sequence modeling, which supports three computation paradigms, i.e., parallel, recurrent, and chunkwise recurrent. Specifically, the parallel representation allows for training parallelism. The recurrent representation enables low-cost $O(1)$ inference, which improves decoding throughput, latency, and GPU memory without sacrificing performance. The chunkwise recurrent representation facilitates efficient long-sequence modeling with linear complexity, where each chunk is encoded parallelly while recurrently summarizing the chunks. Experimental results on language modeling show that RetNet achieves favorable scaling results, parallel training, low-cost deployment, and efficient inference. The intriguing properties make RetNet a strong successor to Transformer for large language models. Code will be available at https://aka.ms/retnet.