Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/thuml/Flowformer
About Code release for "Flowformer: Linearizing Transformers with Conservation Flows" (ICML 2022), https://arxiv.org/pdf/2202.06258.pdf
https://github.com/thuml/Flowformer
deep-learning
Last synced: about 1 month ago
JSON representation
About Code release for "Flowformer: Linearizing Transformers with Conservation Flows" (ICML 2022), https://arxiv.org/pdf/2202.06258.pdf
- Host: GitHub
- URL: https://github.com/thuml/Flowformer
- Owner: thuml
- License: mit
- Created: 2022-06-07T08:45:10.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-07-18T02:34:14.000Z (5 months ago)
- Last Synced: 2024-08-03T09:09:30.499Z (5 months ago)
- Topics: deep-learning
- Language: Python
- Homepage:
- Size: 6.36 MB
- Stars: 298
- Watchers: 9
- Forks: 28
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- StarryDivineSky - thuml/Flowformer
README
# Flowformer (ICML 2022)
Flowformer: Linearizing Transformers with Conservation Flows
:triangular_flag_on_post:**News** (2024.07) [Mobile-Attention](https://openreview.net/pdf?id=VHtIDVaOKC), a mobile-device-tailored version of Flowformer, has been published in ICML 2024. The attention code can be found [here](https://github.com/thuml/MobileAttention/blob/main/Mobile_Attention.py). You can obtain a faster model by just replacing the canonical Attention Mechanism with our Mobile-Attention.
Transformers have achieved impressive success in various areas. However, the attention mechanism has quadratic complexity, significantly impeding Transformers from dealing with numerous tokens and scaling up to bigger models. In pursuing the **linear complexity** and **task-universal** foundation model, we propose Flowformer [[paper]](https://arxiv.org/pdf/2202.06258.pdf) with the following merits:
- **Linear complexity** w.r.t sequence length, can handle extremely long sequences (over 4k tokens)
- **Without specific inductive bias**, purely derived from the flow network theory
- **Task-universal**, showing strong performance in **$\color{red}{\text{Long sequence, Vision, NLP, Time series, RL}}$**.
## Flow-Attention Design
We cast the attention mechanism into flow network, where the information flow is aggregated from the sources (values) to the sinks (results) through the learned flow capacities (attentions).
By conducting the conservation in both source and sink ascpects, we can bring competition into Flow-Attention design to avoid trivial attention in the spirit that "fixed resource will cause competition''.
Figure 1. Flow-Attention with Competition and Allocation mechanisms.
## Get Started
1. Please refer to different folders for detailed experiment instructions.
Note: We have suffered a lot in configuring environments for different tasks. If you also have problems in solving the environment, feel free to contact us and discuss about it.
2. List of benchmarks
- [x] Core code: see `Flow_Attention.py`
- [x] GPT-style Pytorch Module: see `Flowformer_TorchModule`
- [x] Long Sequence Modeling in LRA: see `Flowformer_LRA`
- [x] Vision Recognization in ImageNet-1K: see `Flowformer_CV`
- [x] Language Modeling in WikiText-103: see `Flowformer_NLP`
- [x] Time series classification in UEA: see `Flowformer_TimeSeries`
- [x] Reinforcement Learning in D4RL: see `Flowformer_RL`
- [ ] CUDA speed up version
## Main Results
See the [[paper]](https://arxiv.org/pdf/2202.06258.pdf) for detailed results, including **nearly 20 comparing baselines**.
| Task | Metrics | Flowformer | Performer | Reformer | Vanilla
Transformer |
| ------------------------------------- | ------------------------------------------------ | ---------------- | ------------ | ------------ | ---------------------- |
| Long Sequence Modeling
(LRA) | Avg Acc (%) $\uparrow$ | **56.48** | 51.41 | 50.67 | OOM |
| Vision Recognization
(ImageNet-1K) | Top-1 Acc (%) $\uparrow$ | **80.6** | 78.1 | 79.6 | 78.7 |
| Language Modeling
(WikiText-103) | Perplexity $\downarrow$ | **30.8** | 37.5 | 33.6 | 33.0 |
| Time series classification
(UEA) | Avg Acc (%) $\uparrow$ | **73.0** | 71.5 | 71.9 | 71.9 |
| Offline RL
(D4RL) | Avg Reward $\uparrow$
Avg Deviation $\downarrow$ | **73.5** $\pm$ 2.9 | 63.8 $\pm$ 7.6 | 63.9 $\pm$ 2.9 | 72.2 $\pm$ **2.6** |
Vanilla Transformer means Decision Transorfomer in RL.
## Attention Visualization
Figure 2. Attention visualization. Flowformer can capture the essential parts successfully.
## Citation
If you find this repo useful, please cite our paper.
```
@inproceedings{wu2022flowformer,
title={Flowformer: Linearizing Transformers with Conservation Flows},
author={Haixu Wu and Jialong Wu and Jiehui Xu and Jianmin Wang and Mingsheng Long},
booktitle={International Conference on Machine Learning},
year={2022}
}
```
## Contact
If you have any questions or want to use the code, please contact [email protected].