RWKV-edge: Deeply Compressed RWKV for Resource-Constrained Devices
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911235658743808 |
|---|---|
| author | Choe, Wonkyo Ji, Yangfeng Lin, Felix Xiaozhu |
| author_facet | Choe, Wonkyo Ji, Yangfeng Lin, Felix Xiaozhu |
| contents | To deploy LLMs on resource-contained platforms such as mobile robots and smartphones, non-transformers LLMs have achieved major breakthroughs. Recently, a novel RNN-based LLM family, Repentance Weighted Key Value (RWKV) has shown strong computational efficiency; nevertheless, RWKV models still have high parameter counts which limited their deployment. In this paper, we propose a suite of compression techniques, ranging from model architecture optimizations to post-training compression, tailored to the RWKV architecture. Combined, our techniques reduce the memory footprint of RWKV models by 3.4x -- 5x with only negligible degradation in accuracy; compared to transformer LLMs with similar accuracy, our models require 4x less memory footprint. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_10856 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | RWKV-edge: Deeply Compressed RWKV for Resource-Constrained Devices Choe, Wonkyo Ji, Yangfeng Lin, Felix Xiaozhu Machine Learning Performance To deploy LLMs on resource-contained platforms such as mobile robots and smartphones, non-transformers LLMs have achieved major breakthroughs. Recently, a novel RNN-based LLM family, Repentance Weighted Key Value (RWKV) has shown strong computational efficiency; nevertheless, RWKV models still have high parameter counts which limited their deployment. In this paper, we propose a suite of compression techniques, ranging from model architecture optimizations to post-training compression, tailored to the RWKV architecture. Combined, our techniques reduce the memory footprint of RWKV models by 3.4x -- 5x with only negligible degradation in accuracy; compared to transformer LLMs with similar accuracy, our models require 4x less memory footprint. |
| title | RWKV-edge: Deeply Compressed RWKV for Resource-Constrained Devices |
| topic | Machine Learning Performance |
| url | https://arxiv.org/abs/2412.10856 |