← All questions
LLM FoundationsHard· SSM· Associative Recall· Model Architecture

What is associative recall, and why did early state-spa...

Associative recall is the task of retrieving a specific value associated with a key after many intervening tokens. Early SSMs struggled with this because they used fixed linear recurrences, causing every token to blend into the state identically. Mamba addresses this through input-dependent discretization, allowing the model to selectively amplify relevant bindings and preserve the query signal.

Code playground (Python)