# HART: Hybrid Attention-RWKV Time series Foundation Model

arXiv: https://arxiv.org/abs/2610.12883 · HTML: https://mlsanity.com/paper/2610.12883/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 42 references that are in the map.
- First posted on arXiv 9 Oct 2026.

## Builds on (most cited first, 22 of 22)

- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [RWKV: Reinventing RNNs for the Transformer Era](https://mlsanity.com/paper/2305.13048/) (2023)
- [A decoder-only foundation model for time-series forecasting](https://mlsanity.com/paper/2310.10688/) (2023)
- [Muon is Scalable for LLM Training](https://mlsanity.com/paper/2502.16982/) (2025)
- [Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free](https://mlsanity.com/paper/2505.06708/) (2025)
- [GIFT-Eval: A Benchmark For General Time Series Forecasting Model Evaluation](https://mlsanity.com/paper/2410.10393/) (2024)
- [Chronos-2: From Univariate to Universal Forecasting](https://mlsanity.com/paper/2510.15821/) (2025)
- [Sundial: A Family of Highly Capable Time Series Foundation Models](https://mlsanity.com/paper/2502.00816/) (2025)
- [Eagle and Finch: RWKV with Matrix-Valued States and Dynamic Recurrence](https://mlsanity.com/paper/2404.05892/) (2024)
- [RWKV-7 "Goose" with Expressive Dynamic State Evolution](https://mlsanity.com/paper/2503.14456/) (2025)
- [TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning](https://mlsanity.com/paper/2505.23719/) (2025)
- [Moirai 2.0: When Less Is More for Time Series Forecasting](https://mlsanity.com/paper/2511.11698/) (2025)
- [From Tables to Time: Extending TabPFN-v2 to Time Series Forecasting](https://mlsanity.com/paper/2501.02945/) (2025)
- [It's TIME: Towards the Next Generation of Time Series Forecasting Benchmarks](https://mlsanity.com/paper/2602.12147/) (2026)
- [Toto 2.0: Time Series Forecasting Enters the Scaling Era](https://mlsanity.com/paper/2605.20119/) (2026)
- [Output Scaling: YingLong-Delayed Chain of Thought in a Large Pretrained Time Series Forecasting Model](https://mlsanity.com/paper/2506.11029/) (2025)
- [xLSTM-Mixer: Multivariate Time Series Forecasting by Mixing via Scalar Memories](https://mlsanity.com/paper/2410.16928/) (2024)
- [CT-PatchTST: Channel-Time Patch Time-Series Transformer for Long-Term Renewable Energy Forecasting](https://mlsanity.com/paper/2501.08620/) (2025)
- [Xihe: Scalable Zero-Shot Time Series Learner Via Hierarchical Interleaved Block Attention](https://mlsanity.com/paper/2510.21795/) (2025)
- [TS-ICL: A Flexible Time-Indexed Foundation Model for Time Series via In-Context Learning](https://mlsanity.com/paper/2606.05878/) (2026)
- [ST-Mamba: Spatial-Temporal Selective State Space Model for Traffic Flow Prediction](https://mlsanity.com/paper/2404.13257/) (2024)
- [Synergizing Global Pattern Learning and Time Order Characterization in Mobile Channel Prediction: An RWKV-Based Approach](https://mlsanity.com/paper/2606.16170/) (2026)

## Abstract (arXiv)

Non-autoregressive multi-step forecasting (NAF) enables time series foundation models to predict an entire future horizon in a single forward pass. Despite its strong sequence modeling capabilities, RWKV-7 exhibits long-horizon degradation in our NAF experiments. In this setup, future patches are represented as missing inputs, requiring the recurrent state to evolve across the prediction horizon without new observations. Our analysis of RWKV-7's generalized delta rule suggests a mechanism we term Placeholder Idling Decay (PID): repeated state transitions can attenuate historical information when placeholder-driven writes provide insufficient compensation. Layer-wise analysis reveals reduced state-injection norms in most layers, while transition-matrix norms remain similar between historical and prediction regions, consistent with PID. Motivated by this analysis, we propose HART, a Hybrid Attention-RWKV Time-series Foundation Model that alternates attention and RWKV blocks, combining direct access to historical tokens with recurrent temporal modeling. Controlled experiments on GIFT-Eval and TIME-Benchmark show that HART improves over pure RWKV at medium and long horizons while maintaining comparable short-horizon performance, and achieves better aggregate performance than a comparably sized pure Transformer. With 36M parameters, HART achieves long-horizon MASE comparable to Toto-2.0-1B on GIFT-Eval (0.752 vs. 0.754), using approximately 1/28 of its parameters.