RWKV-TS

Summary

RWKV-TS is a time-series adaptation of RWKV-style recurrent sequence modeling. It uses patching, time mixing, channel mixing, and WKV recurrence to target competitive performance with lower sequence-length cost than quadratic attention.

Role In The Wiki

RWKV-TS is the direct time-series bridge for efficient recurrent sequence models. It should be compared with xLSTM, SSM, Mamba, and ParaRNN-style paths, while keeping its trained-from-scratch evaluation separate from pretrained TSFM leaderboards.

Official Artifacts

• Official code: https://github.com/howard-hou/RWKV-TS

Evidence

• RWKV-TS: Beyond Traditional Recurrent Neural Network for Time Series Tasks

Relation To Foundation TSFM Agenda

Use the source-level agenda mapping in rwkv-ts-2024 rather than duplicating verdict rows here.

At the entity level, RWKV-TS is the direct time-series bridge for efficient recurrent sequence models. It should be compared with xLSTM, SSM, Mamba, and ParaRNN-style paths, while keeping its trained-from-scratch evaluation separate from pretrained TSFM leaderboards. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.

Related Pages

• Foundation Time-Series Model Research Agenda
• Efficient Recurrent Sequence Models
• Time-Series Scaling And Efficiency
• Time-Series Foundation Models
• Mamba
• Mamba-2
• Mamba-3
• ParaRNN
• Recurrent Memory Transformer