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
Ultra-Data-Efficient GAN Training: Drawing A Lottery Ticket First, Then Training It Toughly

Ultra-Data-Efficient GAN Training: Drawing A Lottery Ticket First, Then Training It Toughly Code for this paper Ultra-Data-Efficient GAN Tra

VITA 77 Oct 05, 2022
Implementation for paper MLP-Mixer: An all-MLP Architecture for Vision

MLP Mixer Implementation for paper MLP-Mixer: An all-MLP Architecture for Vision. Give us a star if you like this repo. Author: Github: bangoc123 Emai

Ngoc Nguyen Ba 86 Dec 10, 2022
Arxiv harvester - Poor man's simple harvester for arXiv resources

Poor man's simple harvester for arXiv resources This modest Python script takes

Patrice Lopez 5 Oct 18, 2022
The tl;dr on a few notable transformer/language model papers + other papers (alignment, memorization, etc).

The tl;dr on a few notable transformer/language model papers + other papers (alignment, memorization, etc).

Will Thompson 166 Jan 04, 2023
Sleep staging from ECG, assisted with EEG

Sleep_Staging_Knowledge Distillation This codebase implements knowledge distillation approach for ECG based sleep staging assisted by EEG based sleep

2 Dec 12, 2022
A deep learning model for style-specific music generation.

DeepJ: A model for style-specific music generation https://arxiv.org/abs/1801.00887 Abstract Recent advances in deep neural networks have enabled algo

Henry Mao 704 Nov 23, 2022
Code and data of the Fine-Grained R2R Dataset proposed in paper Sub-Instruction Aware Vision-and-Language Navigation

Fine-Grained R2R Code and data of the Fine-Grained R2R Dataset proposed in the EMNLP2020 paper Sub-Instruction Aware Vision-and-Language Navigation. C

YicongHong 34 Nov 15, 2022
Uncertain natural language inference

Uncertain Natural Language Inference This repository hosts the code for the following paper: Tongfei Chen*, Zhengping Jiang*, Adam Poliak, Keisuke Sak

Tongfei Chen 14 Sep 01, 2022
Classification Modeling: Probability of Default

Credit Risk Modeling in Python Introduction: If you've ever applied for a credit card or loan, you know that financial firms process your information

Aktham Momani 2 Nov 07, 2022
Chinese clinical named entity recognition using pre-trained BERT model

Chinese clinical named entity recognition (CNER) using pre-trained BERT model Introduction Code for paper Chinese clinical named entity recognition wi

Xiangyang Li 109 Dec 14, 2022
Turn based roguelike in python

pyTB Turn based roguelike in python Documentation can be found here: http://mcgillij.github.io/pyTB/index.html Screenshot Dependencies Written in Pyth

Jason McGillivray 4 Sep 29, 2022
Wandb-predictions - WANDB Predictions With Python

WANDB API CI/CD Below we capture the CI/CD scenarios that we would expect with o

Anish Shah 6 Oct 07, 2022
Code for the IJCAI 2021 paper "Structure Guided Lane Detection"

SGNet Project for the IJCAI 2021 paper "Structure Guided Lane Detection" Abstract Recently, lane detection has made great progress with the rapid deve

Jinming Su 27 Dec 08, 2022
NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset

NOD (Night Object Detection) Dataset NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset, BM

Igor Morawski 17 Nov 05, 2022
Generative Query Network (GQN) in PyTorch as described in "Neural Scene Representation and Rendering"

Update 2019/06/24: A model trained on 10% of the Shepard-Metzler dataset has been added, the following notebook explains the main features of this mod

Jesper Wohlert 313 Dec 27, 2022
Open Source Differentiable Computer Vision Library for PyTorch

Kornia is a differentiable computer vision library for PyTorch. It consists of a set of routines and differentiable modules to solve generic computer

kornia 7.6k Jan 04, 2023
I tried to apply the CAM algorithm to YOLOv4 and it worked.

YOLOV4：You Only Look Once目标检测模型在pytorch当中的实现 2021年2月7日更新： 加入letterbox_image的选项，关闭letterbox_image后网络的map得到大幅度提升。 目录 性能情况 Performance 实现的内容 Achievement

55 Dec 05, 2022
Pytorch implement of 'Unmixing based PAN guided fusion network for hyperspectral imagery'

Pgnet There's a improved version compared with the publication in Tgrs with the modification in the deduction of the PDIN block: https://arxiv.org/abs

5 Jul 01, 2022
Devkit for 3D -- Some utils for 3D object detection based on Numpy and Pytorch

D3D Devkit for 3D: Some utils for 3D object detection and tracking based on Numpy and Pytorch Please consider siting my work if you find this library

Jacob Zhong 27 Jul 07, 2022
CS506-Spring2022 - Code and Slides for Boston University CS 506

CS 506 - Computational Tools for Data Science Code, slides, and notes for Boston

Lance Galletti 17 May 06, 2022