Key-Value Transformer

Transformers have emerged as the prevailing standard solution for various AI tasks, including computer vision and natural language processing. The widely adopted Query, Key, and Value formulation (Q<PRE_TAG>KV</POST_TAG>) has played a significant role in this. Nevertheless, no research has examined the essentiality of these three components for transformer performance. Therefore, we conducted an evaluation of the key-value formulation (KV), which generates symmetric attention maps, along with an asymmetric version that incorporates a 2D positional encoding into the attention matrix. Remarkably, this transformer requires fewer parameters and computation than the original one. Through experiments encompassing three task types -- synthetics (such as reversing or sorting a list), vision (mnist or cifar classification), and NLP (character generation and translation) -- we discovered that the KV transformer occasionally outperforms the Q<PRE_TAG>KV</POST_TAG> transformer. However, it also exhibits instances of underperformance compared to Q<PRE_TAG>KV</POST_TAG>, making it challenging to draw a definitive conclusion. Nonetheless, we consider the reported results to be encouraging and anticipate that they may pave the way for more efficient transformers in the future.