Millions of States: Designing a Scalable MoE Architecture with RWKV-7 Meta-learner
This work addresses scalability and adaptability issues in state-based sequence models for efficient sequence modeling, though it appears incremental as it builds directly on RWKV-7.
The paper tackles the limitations of RWKV-7, such as lack of token-parameter interactions and scalability, by proposing Meta-State, a novel extension that integrates these features through a Self-State Encoder mechanism, enabling progressive model scaling without retraining while maintaining linear complexity and constant memory usage.
State-based sequence models like RWKV-7 offer a compelling alternative to Transformer architectures, achieving linear complexity while demonstrating greater expressive power in short-context scenarios and enabling state tracking beyond the \(\text{TC}^0\) complexity class. However, RWKV-7 lacks mechanisms for token-parameter interactions and native scalability, limiting its adaptability and growth without retraining. In this paper, we propose \textbf{Meta-State}, a novel extension to RWKV-7 that replaces attention mechanisms with a fully state-driven approach, integrating token-parameter interactions through a \textbf{Self-State Encoder} (SSE) mechanism. The SSE repurposes a portion of the RWKV-7 Weighted Key-Value (WKV) state as transformation weights to encode token-parameter interactions in a linear, state-driven manner without introducing new trainable matrices or softmax operations, while preserving the autoregressive property of token processing. Meta-State supports progressive model scaling by expanding the WKV state and parameter tokens, reusing existing parameters without retraining. Our approach bridges the gap between state-based modeling, token-parameter interactions, and scalable architectures, offering a flexible framework for efficient and adaptable sequence modeling with linear complexity and constant memory usage.