Implementation of Memformer, a Memory-augmented Transformer, in Pytorch

Overview

Memformer - Pytorch

Implementation of Memformer, a Memory-augmented Transformer, in Pytorch. It includes memory slots, which are updated with attention, learned efficiently through Memory-Replay BackPropagation (MRBP) through time.

Install

$ pip install memformer

Usage

Full encoder / decoder, as in the paper

import torch
from memformer import Memformer

model = Memformer(
    dim = 512,
    enc_num_tokens = 256,
    enc_depth = 2,
    enc_heads = 8,
    enc_max_seq_len = 1024,
    dec_num_tokens = 256,
    dec_depth = 2,
    dec_heads = 8,
    dec_max_seq_len = 1024,
    num_memory_slots = 128
)

src_seg_1 = torch.randint(0, 256, (1, 1024))
src_seg_2 = torch.randint(0, 256, (1, 1024))
src_seg_3 = torch.randint(0, 256, (1, 1024))

tgt = torch.randint(0, 256, (1, 1024))

enc_out1, mems1,    _ = model(src_seg_1) # (1, 1024, 512), (1, 128, 512), _
enc_out2, mems2,    _ = model(src_seg_2, mems = mems1)
enc_out3, mems3, loss = model(src_seg_3, tgt, mems = mems2)

loss.backward()

Encoder only

import torch
from memformer import Memformer

model = Memformer(
    dim = 512,
    enc_num_tokens = 256,
    enc_heads = 8,
    enc_depth = 2,
    enc_max_seq_len = 1024,
    num_memory_slots = 128,
    num_mem_updates = 2,
    encoder_only = True       # only use encoder, in which output is encoded output
)

src1 = torch.randint(0, 256, (1, 1024))
src2 = torch.randint(0, 256, (1, 1024))

enc1, mems1 = model(src1) # (1, 1024, 512), (1, 128, 512)
enc2, mems2 = model(src2, mems = mems1)

Memory Replay Back-Propagation

import torch
from memformer import Memformer, memory_replay_backprop

model = Memformer(
    dim = 512,
    num_memory_slots = 128,
    enc_num_tokens = 256,
    enc_depth = 2,
    enc_max_seq_len = 1024,
    dec_num_tokens = 256,
    dec_depth = 2,
    dec_max_seq_len = 1024
).cuda()

seq = torch.randint(0, 256, (1, 8192)).cuda()
seq_mask = torch.ones_like(seq).bool().cuda()

tgt = torch.randint(0, 256, (1, 512)).cuda()
tgt_mask = torch.ones_like(tgt).bool().cuda()

# will automatically split the source sequence to 8 segments
memory_replay_backprop(
    model,
    src = seq,
    tgt = tgt,
    src_mask = seq_mask,
    tgt_mask = tgt_mask
)

Citations

@inproceedings{
    anonymous2021memformer,
    title={Memformer: The Memory-Augmented Transformer},
    author={Anonymous},
    booktitle={Submitted to International Conference on Learning Representations},
    year={2021},
    url={https://openreview.net/forum?id=_adSMszz_g9},
    note={under review}
}
You might also like...
Implementation of Vision Transformer, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch
Implementation of Vision Transformer, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch

Implementation of Vision Transformer, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch

Styled Augmented Translation
Styled Augmented Translation

SAT Style Augmented Translation Introduction By collecting high-quality data, we were able to train a model that outperforms Google Translate on 6 dif

TANL: Structured Prediction as Translation between Augmented Natural Languages

TANL: Structured Prediction as Translation between Augmented Natural Languages Code for the paper "Structured Prediction as Translation between Augmen

A neuroanatomy-based augmented reality experience powered by computer vision. Features 3D visuals of the Atlas Brain Map slices.

Brain Augmented Reality (AR) A neuroanatomy-based augmented reality experience powered by computer vision that features 3D visuals of the Atlas Brain

Motion Planner Augmented Reinforcement Learning for Robot Manipulation in Obstructed Environments (CoRL 2020)
Motion Planner Augmented Reinforcement Learning for Robot Manipulation in Obstructed Environments (CoRL 2020)

Motion Planner Augmented Reinforcement Learning for Robot Manipulation in Obstructed Environments [Project website] [Paper] This project is a PyTorch

A heterogeneous entity-augmented academic language model based on Open Academic Graph (OAG)
A heterogeneous entity-augmented academic language model based on Open Academic Graph (OAG)

Library | Paper | Slack We released two versions of OAG-BERT in CogDL package. OAG-BERT is a heterogeneous entity-augmented academic language model wh

DrQ-v2: Improved Data-Augmented Reinforcement Learning
DrQ-v2: Improved Data-Augmented Reinforcement Learning

DrQ-v2: Improved Data-Augmented RL Agent Method DrQ-v2 is a model-free off-policy algorithm for image-based continuous control. DrQ-v2 builds on DrQ,

[EMNLP 2021] Distantly-Supervised Named Entity Recognition with Noise-Robust Learning and Language Model Augmented Self-Training

RoSTER The source code used for Distantly-Supervised Named Entity Recognition with Noise-Robust Learning and Language Model Augmented Self-Training, p

 RNG-KBQA: Generation Augmented Iterative Ranking for Knowledge Base Question Answering
RNG-KBQA: Generation Augmented Iterative Ranking for Knowledge Base Question Answering

RNG-KBQA: Generation Augmented Iterative Ranking for Knowledge Base Question Answering Authors: Xi Ye, Semih Yavuz, Kazuma Hashimoto, Yingbo Zhou and

Comments
  • WIP - MemformerEncoder

    WIP - MemformerEncoder

    I´m always trying all your awesome work on transformers. My problem is NER on very large texts, with few examples.

    Memformer is the first one so far to converge faster and wield better accuracy than RNN encoders as LSTM, SRU and IndRNN It is ridiculously better than everything else I tested, congratulations @lucidrains 🥳

    I need to use the transformer as a Encoder in my pipeline, to feed a CRF layer. So I modified the code to accept an already embedded input, and to only do the Encode step.

    TODO:

    • [ ] Support Mask
    • [ ] Re-utilize code with Memformer class

    Is this within the scope of the project?

    opened by bratao 10
  • ETA on complete examples

    ETA on complete examples

    @lucidrains As I asked about the feedback-transformer, I was also wondering about this memformer implementation as I would love to try it. Any eta on any complete examples here? They will be much appreciated. Thanks.

    And similarly, I would love to see a simple example for custom line-by-line TXT datasets as well.

    Thank you again :)

    opened by asigalov61 0
Owner
Phil Wang
Working with Attention. It's all we need
Phil Wang
Underwater image enhancement

LANet Our work proposes an adaptive learning attention network (LANet) to solve the problem of color casts and low illumination in underwater images.

LiuShiBen 7 Sep 14, 2022
The code release of paper Low-Light Image Enhancement with Normalizing Flow

[AAAI 2022] Low-Light Image Enhancement with Normalizing Flow Paper | Project Page Low-Light Image Enhancement with Normalizing Flow Yufei Wang, Renji

Yufei Wang 176 Jan 06, 2023
zeus is a Python implementation of the Ensemble Slice Sampling method.

zeus is a Python implementation of the Ensemble Slice Sampling method. Fast & Robust Bayesian Inference, Efficient Markov Chain Monte Carlo (MCMC), Bl

Minas Karamanis 197 Dec 04, 2022
A PyTorch Lightning solution to training OpenAI's CLIP from scratch.

train-CLIP 📎 A PyTorch Lightning solution to training CLIP from scratch. Goal ⚽ Our aim is to create an easy to use Lightning implementation of OpenA

Cade Gordon 396 Dec 30, 2022
TraSw for FairMOT - A Single-Target Attack example (Attack ID: 19; Screener ID: 24):

TraSw for FairMOT A Single-Target Attack example (Attack ID: 19; Screener ID: 24): Fig.1 Original Fig.2 Attacked By perturbing only two frames in this

Derry Lin 21 Dec 21, 2022
Tools for robust generative diffeomorphic slice to volume reconstruction

RGDSVR Tools for Robust Generative Diffeomorphic Slice to Volume Reconstructions (RGDSVR) This repository provides tools to implement the methods in t

Lucilio Cordero-Grande 0 Oct 29, 2021
Reference models and tools for Cloud TPUs.

Cloud TPUs This repository is a collection of reference models and tools used with Cloud TPUs. The fastest way to get started training a model on a Cl

5k Jan 05, 2023
The code for Bi-Mix: Bidirectional Mixing for Domain Adaptive Nighttime Semantic Segmentation

BiMix The code for Bi-Mix: Bidirectional Mixing for Domain Adaptive Nighttime Semantic Segmentation arxiv Framework: visualization results: Requiremen

stanley 18 Sep 18, 2022
CVPR 2021 - Official code repository for the paper: On Self-Contact and Human Pose.

TUCH This repo is part of our project: On Self-Contact and Human Pose. [Project Page] [Paper] [MPI Project Page] License Software Copyright License fo

Lea Müller 45 Jan 07, 2023
DeepI2I: Enabling Deep Hierarchical Image-to-Image Translation by Transferring from GANs

DeepI2I: Enabling Deep Hierarchical Image-to-Image Translation by Transferring from GANs Abstract: Image-to-image translation has recently achieved re

yaxingwang 23 Apr 14, 2022
Manifold-Mixup implementation for fastai V2

Manifold Mixup Unofficial implementation of ManifoldMixup (Proceedings of ICML 19) for fast.ai (V2) based on Shivam Saboo's pytorch implementation of

Nestor Demeure 16 Jul 25, 2022
Dynamic wallpaper generator.

Wiki • About • Installation About This project is a dynamic wallpaper changer. It waits untill you turn on the music, downloads album cover if it's po

3 Sep 18, 2021
Hysterese plugin with two temperature offset areas

craftbeerpi4 plugin OffsetHysterese Temperatur-Steuerungs-Plugin mit zwei tempereaturbereich abhängigen Offsets. Installation sudo pip3 install https:

HappyHibo 1 Dec 21, 2021
This is the code for Compressing BERT: Studying the Effects of Weight Pruning on Transfer Learning

This is the code for Compressing BERT: Studying the Effects of Weight Pruning on Transfer Learning It includes /bert, which is the original BERT repos

Mitchell Gordon 11 Nov 15, 2022
Official implementation for the paper: "Multi-label Classification with Partial Annotations using Class-aware Selective Loss"

Multi-label Classification with Partial Annotations using Class-aware Selective Loss Paper | Pretrained models Official PyTorch Implementation Emanuel

99 Dec 27, 2022
The pytorch implementation of DG-Font: Deformable Generative Networks for Unsupervised Font Generation

DG-Font: Deformable Generative Networks for Unsupervised Font Generation The source code for 'DG-Font: Deformable Generative Networks for Unsupervised

130 Dec 05, 2022
A Unified Generative Framework for Various NER Subtasks.

This is the code for ACL-ICJNLP2021 paper A Unified Generative Framework for Various NER Subtasks. Install the package in the requirements.txt, then u

177 Jan 05, 2023
Toontown: Galaxy, a new Toontown game based on Disney's Toontown Online

Toontown: Galaxy The official archive repo for Toontown: Galaxy, a new Toontown

1 Feb 15, 2022
Json2Xml tool will help you convert from json COCO format to VOC xml format in Object Detection Problem.

JSON 2 XML All codes assume running from root directory. Please update the sys path at the beginning of the codes before running. Over View Json2Xml t

Nguyễn Trường Lâu 6 Aug 22, 2022
Audio2Face - Audio To Face With Python

Audio2Face Discription We create a project that transforms audio to blendshape w

FACEGOOD 724 Dec 26, 2022