Implementation of the Transformer variant proposed in "Transformer Quality in Linear Time"

Related tags

Deep Learningdeep-learningtransformersartificial-intelligenceattention-mechanismefficient-transformers
Overview

FLASH - Pytorch

Implementation of the Transformer variant proposed in the paper Transformer Quality in Linear Time

Install

$ pip install FLASH-pytorch

Usage

The main novel circuit in this paper is the "Gated Attention Unit", which they claim can replace multi-headed attention while reducing it to just one head.

It uses a relu squared activation in place of the softmax, the activation of which was first seen in the Primer paper, and the use of ReLU in ReLA Transformer. The gating style seems mostly inspired by gMLPs.

import torch
from flash_pytorch import GAU

gau = GAU(
    dim = 512,
    query_key_dim = 128,     # query / key dimension
    causal = True,           # autoregressive or not
    expansion_factor = 2,    # hidden dimension = dim * expansion_factor
)

x = torch.randn(1, 1024, 512)
out = gau(x) # (1, 1024, 512)

The authors then combine GAU with Katharopoulos linear attention, using grouping of the sequences to overcome a known issue with autoregressive linear attention.

This combination of the quadratic gated attention unit with grouped linear attention they named FLASH

You can also use this quite easily

import torch
from flash_pytorch import FLASH

flash = FLASH(
    dim = 512,
    group_size = 256,             # group size
    causal = True,                # autoregressive or not
    query_key_dim = 128,          # query / key dimension
    expansion_factor = 2.         # hidden dimension = dim * expansion_factor
)

x = torch.randn(1, 1111, 512)     # sequence will be auto-padded to nearest group size
out = flash(x) # (1, 1111, 512)

Finally, you can use the full FLASH transformer as mentioned in the paper. This contains all the positional embeddings mentioned in the paper. Absolute positional embedding uses scaled sinusoidal. GAU quadratic attention will get one-headed T5 relative positional bias. On top of all this, both GAU attention as well as the linear attention will be rotary embedded (RoPE).

import torch
from flash_pytorch import FLASHTransformer

model = FLASHTransformer(
    num_tokens = 20000,          # number of tokens
    dim = 512,                   # model dimension
    depth = 12,                  # depth
    causal = True,               # autoregressive or not
    group_size = 256,            # size of the groups
    query_key_dim = 128,         # dimension of queries / keys
    expansion_factor = 2.,       # hidden dimension = dim * expansion_factor
    norm_type = 'scalenorm',     # in the paper, they claimed scalenorm led to faster training at no performance hit. the other option is 'layernorm' (also default)
    shift_tokens = True          # discovered by an independent researcher in Shenzhen @BlinkDL, this simply shifts half of the feature space forward one step along the sequence dimension - greatly improved convergence even more in my local experiments
)

x = torch.randint(0, 20000, (1, 1024))
logits = model(x) # (1, 1024, 20000)

Test on Autoregressive Enwik8

$ python train.py

Citations

@article{Hua2022TransformerQI,
    title   = {Transformer Quality in Linear Time},
    author  = {Weizhe Hua and Zihang Dai and Hanxiao Liu and Quoc V. Le},
    journal = {ArXiv},
    year    = {2022},
    volume  = {abs/2202.10447}
}
@software{peng_bo_2021_5196578,
    author    = {PENG Bo},
    title     = {BlinkDL/RWKV-LM: 0.01},
    month     = {aug},
    year      = {2021},
    publisher = {Zenodo},
    version   = {0.01},
    doi       = {10.5281/zenodo.5196578},
    url       = {https://doi.org/10.5281/zenodo.5196578}
}
Comments
  • einsum operation in Linear Attention Part

    einsum operation in Linear Attention Part

    Hi, Thanks a lot for your FLASH_pytorch, which helps a lot. I found that there are some differences from the paper in the Linear Attention Part: https://github.com/lucidrains/FLASH-pytorch/blob/main/flash_pytorch/flash_pytorch.py#L342-L343

    lin_kv = einsum('b g n d, b g n e -> b d e', lin_k, v) / n
lin_out = einsum('b g n d, b d e -> b g n e', lin_q, lin_kv)

    the lin_kv is three-dim (bde) And the code in the paper is

    lin_kv = tf.einsum('bhke,bgh→bgke', lin_kv, mask) 
linear = tf.einsum('bgnk,bgke→bgne', lin_q, lin_kv)

    the lin_kv is four-dim (bgke) It seems that the two ways are not equivalent.

    Looking forward to your reply. Best,

    opened by ShomyLiu 5
  • mask error

    mask error 

    x = torch.randint(0, 20000, (1, 1024))
mask = x.ne(0)
logits = model(x, mask=mask)

    RuntimeError: The size of tensor a (1024) must match the size of tensor b (128) at non-singleton dimension 2

    opened by keyunluo 1
  • Speed on TPU

    Speed on TPU

    Hi, Thanks for the code! I test it on Google TPU v3, the training speed seems slower than my expectation. Maybe there is some operation which is not lower on TPU.

    opened by magicknight 0
  • About the

    About the "shift_tokens"

    Thank you for your amazing code.

    In the class of FLASH, I find a flag: shift_tokens, and the corresponding code is as following: if self.shift_tokens: x_shift, x_pass = normed_x.chunk(2, dim = -1) x_shift = F.pad(x_shift, (0, 0, 1, -1), value = 0.) normed_x = torch.cat((x_shift, x_pass), dim = -1)

    Assume we have normed_x in the shape [1024, 512], the x_shift/x_pass is the shape of [1024, 256]. Then it adds a row (with all 0 value) and remove the last row in the x_shift, and concat x_shift and x_pass to get the normed_x.

    In my opinion, the F.pad operation will make the row in x_shift and x_pass do not match again.

    May I know why it works?

    Kang

    opened by kangzhao2 1
  • Cross-Attention?

    Cross-Attention?

    Hi, @lucidrains. Thank you for sharing this excellent implementation with us all! Do you have any thoughts as to what changes would need to be made to make cross-attention possible with your FLASH model?

    opened by amorehead 2
Owner
Phil Wang
Working with Attention. It's all we need
Phil Wang
GitHub Repository
This repository contains the source code for the paper First Order Motion Model for Image Animation

!!! Check out our new paper and framework improved for articulated objects First Order Motion Model for Image Animation This repository contains the s

13k Jan 09, 2023
PyTorch implementation of Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets

Simple PyTorch Implementation of "Grokking" Implementation of Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets Usage Running

Teddy Koker 15 Sep 29, 2022
Analysis of Smiles through reservoir sampling & RDkit

Analysis of Smiles through reservoir sampling and machine learning (under development). This is a simple project that includes two Jupyter files for t

Aurimas A. Nausėdas 6 Aug 30, 2022
Social Distancing Detector

Computer vision has opened up a lot of opportunities to explore into AI domain that were earlier highly limited. Here is an application of haarcascade classifier and OpenCV to develop a social distan

Ashish Pandey 2 Jul 18, 2022
This repo contains the pytorch implementation for Dynamic Concept Learner (accepted by ICLR 2021).

DCL-PyTorch Pytorch implementation for the Dynamic Concept Learner (DCL). More details can be found at the project page. Framework Grounding Physical

Zhenfang Chen 31 Jan 06, 2023
Bi-level feature alignment for versatile image translation and manipulation (Under submission of TPAMI)

Bi-level feature alignment for versatile image translation and manipulation (Under submission of TPAMI) Preparation Clone the Synchronized-BatchNorm-P

Fangneng Zhan 12 Aug 10, 2022
Official code for the publication "HyFactor: Hydrogen-count labelled graph-based defactorization Autoencoder".

HyFactor Graph-based architectures are becoming increasingly popular as a tool for structure generation. Here, we introduce a novel open-source archit

Laboratoire-de-Chemoinformatique 11 Oct 10, 2022
automated systems to assist guarding corona Virus precautions for Closed Rooms (e.g. Halls, offices, etc..)

Automatic-precautionary-guard automated systems to assist guarding corona Virus precautions for Closed Rooms (e.g. Halls, offices, etc..) what is this

badra 0 Jan 06, 2022
A Closer Look at Structured Pruning for Neural Network Compression

A Closer Look at Structured Pruning for Neural Network Compression Code used to reproduce experiments in https://arxiv.org/abs/1810.04622. To prune, w

Bayesian and Neural Systems Group 140 Dec 05, 2022
Exploring the Dual-task Correlation for Pose Guided Person Image Generation

Dual-task Pose Transformer Network The source code for our paper "Exploring Dual-task Correlation for Pose Guided Person Image Generation“ (CVPR2022)

63 Dec 15, 2022
Colar: Effective and Efficient Online Action Detection by Consulting Exemplars, CVPR 2022.

Colar: Effective and Efficient Online Action Detection by Consulting Exemplars This repository is the official implementation of Colar. In this work,

LeYang 246 Dec 13, 2022
Neural network for recognizing the gender of people in photos

Neural Network For Gender Recognition How to test it? Install requirements.txt file using pip install -r requirements.txt command Run nn.py using pyth

Valery Chapman 1 Sep 18, 2022
Discovering Interpretable GAN Controls [NeurIPS 2020]

GANSpace: Discovering Interpretable GAN Controls Figure 1: Sequences of image edits performed using control discovered with our method, applied to thr

Erik Härkönen 1.7k Jan 03, 2023
Python scripts to detect faces in Python with the BlazeFace Tensorflow Lite models

Python scripts to detect faces using Python with the BlazeFace Tensorflow Lite models. Tested on Windows 10, Tensorflow 2.4.0 (Python 3.8).

Ibai Gorordo 46 Nov 17, 2022
public repo for ESTER dataset and modeling (EMNLP'21)

Project / Paper Introduction This is the project repo for our EMNLP'21 paper: https://arxiv.org/abs/2104.08350 Here, we provide brief descriptions of

PlusLab 19 Oct 27, 2022
Combining Reinforcement Learning and Constraint Programming for Combinatorial Optimization

Hybrid solving process for combinatorial optimization problems Combinatorial optimization has found applications in numerous fields, from aerospace to

117 Dec 13, 2022
[CVPRW 21] "BNN - BN = ? Training Binary Neural Networks without Batch Normalization", Tianlong Chen, Zhenyu Zhang, Xu Ouyang, Zechun Liu, Zhiqiang Shen, Zhangyang Wang

BNN - BN = ? Training Binary Neural Networks without Batch Normalization Codes for this paper BNN - BN = ? Training Binary Neural Networks without Bat

VITA 40 Dec 30, 2022
Learning Facial Representations from the Cycle-consistency of Face (ICCV 2021)

Learning Facial Representations from the Cycle-consistency of Face (ICCV 2021) This repository contains the code for our ICCV2021 paper by Jia-Ren Cha

Jia-Ren Chang 40 Dec 27, 2022
YOLOX-CondInst - Implement CondInst which is a instances segmentation method on YOLOX

YOLOX CondInst -- YOLOX 实例分割 前言 本项目是自己学习实例分割时，复现的代码. 通过自己编程，让自己对实例分割有更进一步的了解。 若想

DDGRCF 16 Nov 18, 2022
This repository is the code of the paper Accelerating Deep Reinforcement Learning for Digital Twin Network Optimization with Evolutionary Strategies

ES_OTN_Public Carlos Güemes Palau, Paul Almasan, Pere Barlet Ros, Albert Cabellos Aparicio Contact us: Barcelona Neural Networking Center 8 Oct 11, 2022