Year: 2025
Type: conference-paper
Abstract: State space models (SSMs), such as Mamba, have emerged as an efficient alternative to transformers for long-context sequence modeling.However, despite their growing adoption, SSMs lack the interpretability tools that have been crucial for understanding and improving attention-based architectures.While recent efforts provide insights into Mamba's internal mechanisms, they struggle to capture precise token-level interactions at the layer level, leaving gaps in understanding how Mamba selectively processes sequences across layers.In this work, we introduce LATIM, a novel tokenlevel decomposition method for both Mamba-1 and Mamba-2 that enables fine-grained interpretability.We extensively evaluate our method across diverse tasks, including machine translation, copying, and retrieval-based generation, demonstrating its effectiveness in revealing Mamba's token-to-token interaction patterns.Our code is available at https:// github.com/deep-spin/latim.
Authors: Hugo Pitorro, Marcos Vinicius Treviso
Institution: Instituto de Telecomunicações
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4412945153