StateX framework boosts RNN recall by expanding model states post-training

Researchers have developed StateX, a post-training framework designed to improve the recall capabilities of recurrent neural networks (RNNs). This method efficiently expands the states of pre-trained RNNs, such as linear attention and state-space models, without significantly increasing model parameters. Experiments show StateX enhances recall and in-context learning performance in models up to 1.3 billion parameters, without compromising other functionalities. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enhances recall for RNNs, potentially improving performance on tasks requiring long-context understanding.

RANK_REASON This is a research paper introducing a new framework for improving RNN performance.

Read on arXiv cs.CL →

• StateX
• RNN
• linear attention
• state-space models
• arXiv

• paper
• model release