2Mamba2Furious: Linear in Complexity, Competitive in Accuracy

TL;DR

A method is proposed that is nearly as accurate as softmax attention, yet much more memory efficient for long context lengths, and also investigates elements to Mamba-2 that help surpass softmax attention accuracy.

Abstract

Linear attention transformers have become a strong alternative to softmax attention due to their efficiency. However, linear attention tends to be less expressive and results in reduced accuracy compared to softmax attention. To bridge the accuracy gap between softmax attention and linear attention, we manipulate Mamba-2, a very strong linear attention variant. We first simplify Mamba-2 down to its most fundamental and important components, evaluating which specific choices make it most accurate. From this simplified Mamba variant (Mamba-2S), we improve the A-mask and increase the order of the hidden state, resulting in a method, which we call 2Mamba, that is nearly as accurate as softmax attention, yet much more memory efficient for long context lengths. We also investigate elements to Mamba-2 that help surpass softmax attention accuracy. Code is provided for all our experiments

Figures (11)

• Figure 1: Accuracy of linear attention, Mamba, and softmax attention, keeping everything but the attention mechanism constant across experiments.
• Figure 2: Accuracy of various norm types. Softmax normalization requires a positive inner-product space image, as such we use ReLU.
• Figure 3: Isolated Mamba ablation
• Figure 4: Investigations of major and minor build ups in constructing the simplified Mamba-2 architecture.
• Figure 5: Experimental results comparing accuracy and training stability.