Implementation of Feedback Transformer in Pytorch

Overview

Feedback Transformer - Pytorch

Simple implementation of Feedback Transformer in Pytorch. They improve on Transformer-XL by having each token have access to the representations of all previous layers through time. This is achieved by aggregating the outputs of all layers into a shared memory, which each token across layers can attend to at each time step.

The main drawback is longer training time, due to its non-parallel nature. But I thought I'd build it to further exploration and research into this line of work.

Yannic Kilcher video

I also took the liberty to add some various enhancements, including pre-normalization, GLU gated feedforwards, as well as simplified T5 relative positional embeddings.

Install

$ pip install feedback-transformer-pytorch

Usage

import torch
from feedback_transformer_pytorch import FeedbackTransformer

model = FeedbackTransformer(
    num_tokens = 20000,           # number of tokens
    dim = 512,                    # dimension
    depth = 6,                    # depth
    seq_len = 2,                  # the sequence length of each segment or window
    mem_len = 256,                # length of the memory buffer
    dim_head = 64,                # dimension of each head
    heads = 8,                    # number of heads
    attn_dropout = 0.1,           # attention dropout
    ff_dropout = 0.1              # feedforward dropout
).cuda()

x = torch.randint(0, 20000, (2, 64)).cuda()
model(x)  # (2, 64, 20000)

If you would like to have fine control over the memory (when to detach, etc), you can do it with some extra keyword arguments on .forward

import torch
from feedback_transformer_pytorch import FeedbackTransformer

model = FeedbackTransformer(
    num_tokens = 20000,
    dim = 512,
    depth = 6,
    seq_len = 32,
    mem_len = 256
).cuda()

x1 = torch.randint(0, 20000, (2, 32)).cuda()
x2 = torch.randint(0, 20000, (2, 32)).cuda()
x3 = torch.randint(0, 20000, (2, 32)).cuda()

out1, mem1 = model(x1, return_memory = True)
out2, mem2 = model(x2, memory = mem1, return_memory = True)
out3, mem3 = model(x3, memory = mem2, return_memory = True)  # (2, 32, 20000)

Citations

@misc{fan2021addressing,
    title   = {Addressing Some Limitations of Transformers with Feedback Memory}, 
    author  = {Angela Fan and Thibaut Lavril and Edouard Grave and Armand Joulin and Sainbayar Sukhbaatar},
    year    = {2021},
    eprint  = {2002.09402},
    archivePrefix = {arXiv},
    primaryClass = {cs.LG}
}
Comments
  • Should it really be using lower layers output for keys and values?

    Should it really be using lower layers output for keys and values?

    Could you explain the logic of how the key-value pairs are formed at these lines and whether it is necessary?

    https://github.com/lucidrains/feedback-transformer-pytorch/blob/d7d8939910d1491f01a3d93ce81d4663925fb389/feedback_transformer_pytorch/feedback_transformer_pytorch.py#L146-L151

    It looks to me that line 146 transforms the output of the layer below (x) to keys and values, and the following lines combine these keys and values with the memory. I thought that x should only be used for forming the query here, and only the existing memory is used for keys and values.

    opened by tarvaina 6
  • In place operation with gradient

    In place operation with gradient

    https://github.com/lucidrains/feedback-transformer-pytorch/blob/main/feedback_transformer_pytorch/feedback_transformer_pytorch.py#L173 I think this is an error.

    opened by hadaev8 4
  • Bug in weighted sum

    Bug in weighted sum

    Bug in https://github.com/lucidrains/feedback-transformer-pytorch/blob/main/feedback_transformer_pytorch/feedback_transformer_pytorch.py#L264

    Should be layer_weight = rearrange(layer_weight, 'd -> d () () ()')

    opened by Victor0118 1
  • Input/Output dimensions

    Input/Output dimensions

    Hey @lucidrains

    Can I check the dimensions of the input and output, is it (seq_len, dim) -> (? ,dim, tokens)?

    model = FeedbackTransformer(
        num_tokens = 20000,           # number of tokens
        dim = 512,                    # dimension
        depth = 6,                    # depth
        seq_len = 2,                  # the sequence length of each segment or window
        mem_len = 256,                # length of the memory buffer
        dim_head = 64,                # dimension of each head
        heads = 8,                    # number of heads
        attn_dropout = 0.1,           # attention dropout
        ff_dropout = 0.1              # feedforward dropout
    ).cuda()
    
    x = torch.randint(0, 256, (2, 512)).cuda()
    model(x)  # (1, 512, 20000)
    
    opened by iiSeymour 1
  • Non intuitive memory usage with cross attention

    Non intuitive memory usage with cross attention

    Give simple 256 dim and 512 len tensor and memory len 16 feedback transformer uses 3.6gm memory after forward pass. With cross attention on 100 len tensor usage grows to 14gb.

    While parallel version uses 3.1gb and 3.5gb.

    Notebooks for testing https://colab.research.google.com/drive/1dRImydFn3WthOXdLYIvdf5bsqjXcmhC5?usp=sharing https://colab.research.google.com/drive/1n653j4Pz9_U7OukhTlUbomAHMvpPXwx0?usp=sharing

    opened by hadaev8 0
  • I think mask padding value should be False

    I think mask padding value should be False

    Here https://github.com/lucidrains/feedback-transformer-pytorch/blob/with-cross-attention/feedback_transformer_pytorch/feedback_transformer_pytorch.py#L181

    opened by hadaev8 0
  • ETA for the enwiki8 example

    ETA for the enwiki8 example

    Hey @lucidrains,

    Any eta on the example for auto-regressive enwiki8 example? I and others would really appreciate it as always :)

    Also, if you can provide an example for training on custom line-by-line TXT datasets, it would be absolutely fantastic.

    Thank you.

    opened by asigalov61 0
Owner
Phil Wang
Working with Attention. It's all we need.
Phil Wang
Lightweight Cuda Renderer with Python Wrapper.

pyRender Lightweight Cuda Renderer with Python Wrapper. Compile Change compile.sh line 5 to the glm library include path. This library can be download

Jingwei Huang 53 Dec 02, 2022
This repository implements WGAN_GP.

Image_WGAN_GP This repository implements WGAN_GP. Image_WGAN_GP This repository uses wgan to generate mnist and fashionmnist pictures. Firstly, you ca

Lieon 6 Dec 10, 2021
DeepProbLog is an extension of ProbLog that integrates Probabilistic Logic Programming with deep learning by introducing the neural predicate.

DeepProbLog DeepProbLog is an extension of ProbLog that integrates Probabilistic Logic Programming with deep learning by introducing the neural predic

KU Leuven Machine Learning Research Group 94 Dec 18, 2022
This repository accompanies the ACM TOIS paper "What can I cook with these ingredients?" - Understanding cooking-related information needs in conversational search

In this repository you find data that has been gathered when conducting in-situ experiments in a conversational cooking setting. These data include tr

6 Sep 22, 2022
Explainable Medical ImageSegmentation via GenerativeAdversarial Networks andLayer-wise Relevance Propagation

MedAI: Transparency in Medical Image Segmentation What is this repo This repo contains the code and experiments that are implemented to contribute in

Awadelrahman M. A. Ahmed 1 Nov 22, 2021
Numenta published papers code and data

Numenta research papers code and data This repository contains reproducible code for selected Numenta papers. It is currently under construction and w

Numenta 293 Jan 06, 2023
The authors' official PyTorch SigWGAN implementation

The authors' official PyTorch SigWGAN implementation This repository is the official implementation of [Sig-Wasserstein GANs for Time Series Generatio

9 Jun 16, 2022
3DMV jointly combines RGB color and geometric information to perform 3D semantic segmentation of RGB-D scans.

3DMV 3DMV jointly combines RGB color and geometric information to perform 3D semantic segmentation of RGB-D scans. This work is based on our ECCV'18 p

Владислав Молодцов 0 Feb 06, 2022
HiFi++: a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement

HiFi++ : a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement This is the unofficial implementation of Vocoder part of

Rishikesh (ऋषिकेश) 118 Dec 29, 2022
Checkout some cool self-projects you can try your hands on to curb your boredom this December!

SoC-Winter Checkout some cool self-projects you can try your hands on to curb your boredom this December! These are short projects that you can do you

Web and Coding Club, IIT Bombay 29 Nov 08, 2022
An open framework for Federated Learning.

Welcome to Intel® Open Federated Learning Federated learning is a distributed machine learning approach that enables organizations to collaborate on m

Intel Corporation 397 Dec 27, 2022
BBB streaming without Xorg and Pulseaudio and Chromium and other nonsense (heavily WIP)

BBB Streamer NG? Makes a conference like this... ...streamable like this! I also recorded a small video showing the basic features: https://www.youtub

Lukas Schauer 60 Oct 21, 2022
The official implementation of NeurIPS 2021 paper: Finding Optimal Tangent Points for Reducing Distortions of Hard-label Attacks

The official implementation of NeurIPS 2021 paper: Finding Optimal Tangent Points for Reducing Distortions of Hard-label Attacks

machen 11 Nov 27, 2022
Compact Bidirectional Transformer for Image Captioning

Compact Bidirectional Transformer for Image Captioning Requirements Python 3.8 Pytorch 1.6 lmdb h5py tensorboardX Prepare Data Please use git clone --

YE Zhou 19 Dec 12, 2022
MediaPipe Kullanarak İleri Seviye Bilgisayarla Görü

MediaPipe Kullanarak İleri Seviye Bilgisayarla Görü

Burak Bagatarhan 12 Mar 29, 2022
Official implementation of YOGO for Point-Cloud Processing

You Only Group Once: Efficient Point-Cloud Processing with Token Representation and Relation Inference Module By Chenfeng Xu, Bohan Zhai, Bichen Wu, T

Chenfeng Xu 67 Dec 20, 2022
OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation

Build Type Linux MacOS Windows Build Status OpenPose has represented the first real-time multi-person system to jointly detect human body, hand, facia

25.7k Jan 09, 2023
Freecodecamp Scientific Computing with Python Certification; Solution for Challenge 2: Time Calculator

Assignment Write a function named add_time that takes in two required parameters and one optional parameter: a start time in the 12-hour clock format

Hellen Namulinda 0 Feb 26, 2022
Tensorflow2 Keras-based Semantic Segmentation Models Implementation

Tensorflow2 Keras-based Semantic Segmentation Models Implementation

Hah Min Lew 1 Feb 08, 2022
Build a small, 3 domain internet using Github pages and Wikipedia and construct a crawler to crawl, render, and index.

TechSEO Crawler Build a small, 3 domain internet using Github pages and Wikipedia and construct a crawler to crawl, render, and index. Play with the r

JR Oakes 57 Nov 24, 2022