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

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
Training Confidence-Calibrated Classifier for Detecting Out-of-Distribution Samples / ICLR 2018

Training Confidence-Calibrated Classifier for Detecting Out-of-Distribution Samples This project is for the paper "Training Confidence-Calibrated Clas

168 Nov 29, 2022
Interpretation of T cell states using reference single-cell atlases

Interpretation of T cell states using reference single-cell atlases ProjecTILs is a computational method to project scRNA-seq data into reference sing

Cancer Systems Immunology Lab 139 Jan 03, 2023
🥈78th place in Riiid Answer Correctness Prediction competition

Riiid Answer Correctness Prediction Introduction This repository is the code that placed 78th in Riiid Answer Correctness Prediction competition. Requ

Jungwoo Park 10 Jul 14, 2022
Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning. CVPR 2018

Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning Tensorflow code and models for the paper: Large Scale Fine-Grained Categ

Yin Cui 187 Oct 01, 2022
Complementary Patch for Weakly Supervised Semantic Segmentation, ICCV21 (poster)

CPN (ICCV2021) This is an implementation of Complementary Patch for Weakly Supervised Semantic Segmentation, which is accepted by ICCV2021 poster. Thi

Ferenas 20 Dec 12, 2022
Earth Vision Foundation

EVer - A Library for Earth Vision Researcher EVer is a Pytorch-based Python library to simplify the training and inference of the deep learning model.

Zhuo Zheng 34 Nov 26, 2022
style mixing for animation face

An implementation of StyleGAN on Animation dataset. Install git clone https://github.com/MorvanZhou/anime-StyleGAN cd anime-StyleGAN pip install -r re

Morvan 46 Nov 30, 2022
Multi-task Self-supervised Object Detection via Recycling of Bounding Box Annotations (CVPR, 2019)

Multi-task Self-supervised Object Detection via Recycling of Bounding Box Annotations (CVPR 2019) To make better use of given limited labels, we propo

126 Sep 13, 2022
DeepLab resnet v2 model in pytorch

pytorch-deeplab-resnet DeepLab resnet v2 model implementation in pytorch. The architecture of deepLab-ResNet has been replicated exactly as it is from

Isht Dwivedi 601 Dec 22, 2022
ConE: Cone Embeddings for Multi-Hop Reasoning over Knowledge Graphs

ConE: Cone Embeddings for Multi-Hop Reasoning over Knowledge Graphs This is the code of paper ConE: Cone Embeddings for Multi-Hop Reasoning over Knowl

MIRA Lab 33 Dec 07, 2022
Extract MNIST handwritten digits dataset binary file into bmp images

MNIST-dataset-extractor Extract MNIST handwritten digits dataset binary file into bmp images More info at http://yann.lecun.com/exdb/mnist/ Dependenci

Omar Mostafa 6 May 24, 2021
The 2nd place solution of 2021 google landmark retrieval on kaggle.

Leaderboard, taxonomy, and curated list of few-shot object detection papers.

229 Dec 13, 2022
Fully convolutional deep neural network to remove transparent overlays from images

Fully convolutional deep neural network to remove transparent overlays from images

Marc Belmont 1.1k Jan 06, 2023
Implementation of gaze tracking and demo

Predicting Customer Demand by Using Gaze Detecting and Object Tracking This project is the integration of gaze detecting and object tracking. Predict

2 Oct 20, 2022
This is an example of a reproducible modelling project

An example of a reproducible modelling project What are we doing? This example was created for the 2021 fall lecture series of Stanford's Center for O

Armin Thomas 2 Oct 26, 2021
MinHash, LSH, LSH Forest, Weighted MinHash, HyperLogLog, HyperLogLog++, LSH Ensemble

datasketch: Big Data Looks Small datasketch gives you probabilistic data structures that can process and search very large amount of data super fast,

Eric Zhu 1.9k Jan 07, 2023
利用yolov5和TensorRT从0到1实现目标检测的模型训练到模型部署全过程

写在前面 利用TensorRT加速推理速度是以时间换取精度的做法,意味着在推理速度上升的同时将会有精度的下降,不过不用太担心,精度下降微乎其微。此外,要有NVIDIA显卡,经测试,CUDA10.2可以支持20系列显卡及以下,30系列显卡需要CUDA11.x的支持,并且目前有bug。 默认你已经完成了

Helium 6 Jul 28, 2022
Code for "Searching for Efficient Multi-Stage Vision Transformers"

Searching for Efficient Multi-Stage Vision Transformers This repository contains the official Pytorch implementation of "Searching for Efficient Multi

Yi-Lun Liao 62 Oct 25, 2022
This program automatically runs Python code copied in clipboard

CopyRun This program runs Python code which is copied in clipboard WARNING!! USE AT YOUR OWN RISK! NO GUARANTIES IF ANYTHING GETS BROKEN. DO NOT COPY

vertinski 4 Sep 10, 2021
A curated list of awesome resources related to Semantic Search🔎 and Semantic Similarity tasks.

A curated list of awesome resources related to Semantic Search🔎 and Semantic Similarity tasks.

224 Jan 04, 2023