Eagle and Finch: RWKV with Matrix-Valued States and Dynamic Recurrence
Abstract
We present Eagle (RWKV-5) and Finch (RWKV-6), sequence models improving upon the RWKV (RWKV-4) architecture. Our architectural design advancements include multi-headed matrix-valued states and a dynamic recurrence mechanism that improve expressivity while maintaining the inference efficiency characteristics of RNNs. We introduce a new multilingual corpus with 1.12 trillion tokens and a fast tokenizer based on greedy matching for enhanced multilinguality. We trained four Eagle models, ranging from 0.46 to 7.5 billion parameters, and two Finch models with 1.6 and 3.1 billion parameters and find that they achieve competitive performance across a wide variety of benchmarks. We release all our models on HuggingFace under the Apache 2.0 license. Models at: https://huggingface.co/RWKV Training code at: https://github.com/RWKV/RWKV-LM Inference code at: https://github.com/RWKV/ChatRWKV Time-parallel training code at: https://github.com/RWKV/RWKV-infctx-trainer
Cite
@article{arxiv.2404.05892,
title = {Eagle and Finch: RWKV with Matrix-Valued States and Dynamic Recurrence},
author = {Bo Peng and Daniel Goldstein and Quentin Anthony and Alon Albalak and Eric Alcaide and Stella Biderman and Eugene Cheah and Xingjian Du and Teddy Ferdinan and Haowen Hou and Przemysław Kazienko and Kranthi Kiran GV and Jan Kocoń and Bartłomiej Koptyra and Satyapriya Krishna and Ronald McClelland and Jiaju Lin and Niklas Muennighoff and Fares Obeid and Atsushi Saito and Guangyu Song and Haoqin Tu and Cahya Wirawan and Stanisław Woźniak and Ruichong Zhang and Bingchen Zhao and Qihang Zhao and Peng Zhou and Jian Zhu and Rui-Jie Zhu},
journal= {arXiv preprint arXiv:2404.05892},
year = {2024}
}