Memory Efficient Attention (O(sqrt(n)) for Jax and PyTorch

Overview

Memory Efficient Attention

arXiv PyPI version

This is unofficial implementation of Self-attention Does Not Need O(n^2) Memory for Jax and PyTorch.

Implementation is almost same as the one proposed in the paper, with additional masking and adding bias compatibility, batch dimensions support and PyTorch implementation. For computing attention, the proposed method requires only O(sqrt(n)) memory, and the provided functions can be used as a drop-in replacement for attention calculation.

Important Note: This implementation is a trade-off between memory requirements and runtime, so you should adjust key_chunk_size and query_chunk_size parameters to achieve the best configuration for your usecase. Here is a note from the paper's authors:

While a constant chunk size for the queries and a chunk size of sqrt(n) for the keys and values is optimal for memory consumption, the runtime is also affected by the choice of chunk size in practice, which is heavily affected by the choice of hardware. Ultimately, we have to leave this trade-off to the programmer, and expose the chunk sizes as arguments query_chunk_size and key_chunk_size. In Figure 1 we provide default values for the chunk sizes that lead to minimal runtime impact (on TPUv2), while still providing significant memory savings.

Quick Start

  1. Install the library
# for Jax
pip install memory-efficient-attention[jax]
# for PyTorch
pip install memory-efficient-attention[torch]
# for Running Tests
pip install memory-efficient-attention[testing]
  1. Compute attention with the proper function
0.5 bias = np.random.rand(1, b, 16, 128, 128).astype("float32") / 100 # Adjust chunk sizes efficient_dot_product_attention_jax(query, key, value, mask, bias, key_chunk_size=..., query_chunk_size=...)">
import numpy as np
# for PyTorch
from memory_efficient_attention import efficient_dot_product_attention_pt
# or for Jax
from memory_efficient_attention import efficient_dot_product_attention_jax

# Random Data (batch dimensions are not necessary)
b = 8
query = np.random.rand(1, b, 128, 16, 8).astype("float32")
key = np.random.rand(1, b, 128, 16, 8).astype("float32")
value = np.random.rand(1, b, 128, 16, 8).astype("float32")
# optional, for casual tasks, ...
mask = np.random.rand(1, b, 16, 128, 128) > 0.5
bias = np.random.rand(1, b, 16, 128, 128).astype("float32") / 100

# Adjust chunk sizes        
efficient_dot_product_attention_jax(query, key, value, mask, bias, key_chunk_size=..., query_chunk_size=...)

Citation

Please cite if this implementation helps your research. You can use the following BibTeX entry:

@misc{memory_efficient_attention,
	title = {Memory Efficient Attention},
	author = {Rezaei, Amin},
	howpublished = {\url{github.com/AminRezaei0x443/memory-efficient-attention}},
	year = {2021}
}

Also, for the paper:

@misc{rabe2021selfattention,
      title={Self-attention Does Not Need $O(n^2)$ Memory}, 
      author={Markus N. Rabe and Charles Staats},
      year={2021},
      eprint={2112.05682},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}
You might also like...
Memory-efficient optimum einsum using opt_einsum planning and PyTorch kernels.

opt-einsum-torch There have been many implementations of Einstein's summation. numpy's numpy.einsum is the least efficient one as it only runs in sing

A human-readable PyTorch implementation of "Self-attention Does Not Need O(n^2) Memory"

memory_efficient_attention.pytorch A human-readable PyTorch implementation of "Self-attention Does Not Need O(n^2) Memory" (Rabe&Staats'21). def effic

 Attention for PyTorch with Linear Memory Footprint
Attention for PyTorch with Linear Memory Footprint

Attention for PyTorch with Linear Memory Footprint Unofficially implements https://arxiv.org/abs/2112.05682 to get Linear Memory Cost on Attention (+

Local Attention - Flax module for Jax

Local Attention - Flax Autoregressive Local Attention - Flax module for Jax Install $ pip install local-attention-flax Usage from jax import random fr

Reimplementation of the paper "Attention, Learn to Solve Routing Problems!" in jax/flax.

JAX + Attention Learn To Solve Routing Problems Reinplementation of the paper Attention, Learn to Solve Routing Problems! using Jax and Flax. Fully su

Official and maintained implementation of the paper
Official and maintained implementation of the paper "OSS-Net: Memory Efficient High Resolution Semantic Segmentation of 3D Medical Data" [BMVC 2021].

OSS-Net: Memory Efficient High Resolution Semantic Segmentation of 3D Medical Data Christoph Reich, Tim Prangemeier, Özdemir Cetin & Heinz Koeppl | Pr

Lowest memory consumption and second shortest runtime in NTIRE 2022 challenge on Efficient Super-Resolution

FMEN Lowest memory consumption and second shortest runtime in NTIRE 2022 on Efficient Super-Resolution. Our paper: Fast and Memory-Efficient Network T

Implementation of
Implementation of "Efficient Regional Memory Network for Video Object Segmentation" (Xie et al., CVPR 2021).

RMNet This repository contains the source code for the paper Efficient Regional Memory Network for Video Object Segmentation. Cite this work @inprocee

Rethinking Space-Time Networks with Improved Memory Coverage for Efficient Video Object Segmentation
Rethinking Space-Time Networks with Improved Memory Coverage for Efficient Video Object Segmentation

STCN Rethinking Space-Time Networks with Improved Memory Coverage for Efficient Video Object Segmentation Ho Kei Cheng, Yu-Wing Tai, Chi-Keung Tang [a

Comments
  • feat: output_attentions

    feat: output_attentions

    I'm looking into hacking some of the models in the transformers library to use this library for attention, and I don't see a way to support output_attentions yet. This is a flag passed in transformers, where the attention weights are preserved and returned to the user, if it is set.

    I looked a little at implementing this in the torch backend, and I note the scan() function provides for only a single tensor return value. It seems to me that scan() function would be most clearly replaced by a for loop, but it could also be modified to handle tuples, or return_weights could be handled via accessing nonlocal data in some way instead of returning them through the chunk scanner. I'm also not sure how the output would best be passed to the user.

    Edit: Draft implementation 01/28 at https://github.com/AminRezaei0x443/memory-efficient-attention/compare/main...xloem:faba6371ac7faaa2040a2c26e15ae7ab87f94ce4 . I ended up extending the scan function for parity between implementations. Edit 2: Turns out it's the postsoftmax attention weights, not the presoftmax attention weights. I've updated this post and the draft implementation for this output: https://github.com/AminRezaei0x443/memory-efficient-attention/compare/main...xloem:return_weights

    opened by xloem 4
  • Provide a flag for the user to receive attention weights

    Provide a flag for the user to receive attention weights

    This is my draft code for #1. I saw this feature in the transformers library and wanted to implement it here.

    I'm curious what you think about this feature and implementation.

    The code is simply slightly instrumented so that the final attention weights can be returned to the user. Tests are augmented to test this use. In utils, the scan function is expanded to handle tuples.

    A change to dynamic_slice crept in from dev, to use slices rather than index_slice. I've retained this change because it looks like it would execute faster to me, but it can be removed.

    Rebased and squashed from 84724e1de4721ea0333d6bdbb91e8bce74fbeac .

    opened by xloem 2
  • Improve performance via batched-matmul and fused multiplies

    Improve performance via batched-matmul and fused multiplies

    Many thanks for providing this reference implementation.

    I tried integrating this into stable-diffusion / diffusers. A fix was required to make it work on Mac (PyTorch MPS backend):
    https://github.com/Birch-san/diffusers/pull/1/commits/04372140a25d7f53549175f1f196599c3e9bf3a5

    Knowing that computing attention via baddbmm()+bmm() can outperform einsum by 18%: I tried to rewrite the algorithm to use those.

    I compared the speed of my optimized version, against the implementation in this repository. this result is for "everything fits in one chunk" perf (i.e. chunk size = max token length). I was unable to compare chunked perf, because although I got chunking working in my version: I wasn't able to get it working in the version in this repository (got some unexpected-shape tensors returned).

    compared to the implementation in this repository:
    my optimized version achieves a 2.78x speedup in the time it took to generate a 512x512 image with stable-diffusion v2.1-base (i.e. 4096 vision tokens, 5 attention heads, batch size of 2 due to CFG).

    here's my optimized implementation:
    https://github.com/Birch-san/diffusers/pull/1

    batched matmuls require a 3D tensor, i.e. [batch * num_heads, tokens, channels_per_head].

    code that currently integrates agains this repository's [batch, q_length, num_heads, qk_depth_per_head] format can migrate those tensors to the [batch * num_heads, q_length, channels_per_head] format favoured by my implementation like so:

    query = query.transpose(1,2).flatten(end_dim=1)
    key = key.transpose(1,2).flatten(end_dim=1)
    value = value.transpose(1,2).flatten(end_dim=1)
    

    the result that's returned, remains in [batch * num_heads, q_length, qk_depth_per_head] format, and can be restored to [batch, q_length, num_heads, qk_depth_per_head] format like so:

    result.unflatten(0, (-1, attn.heads)).transpose(1,2)
    

    I think a further speedup is possible too: by working out when chunking is not needed: we can compute whether unchunked attention would fit into memory, and prefer unchunked attention as a fast-path where possible. this will be useful in a Unet, which runs attention at various resolutions.

    EDIT:
    I have now added fast-paths for:

    • skipping kv-chunking when kv_chunk_size >= k_tokens
      • this turns the algorithm into "attention slicing"
    • skipping q-chunking when q_chunk_size >= q_tokens
    • skipping all chunking when the kv_chunk_size >= k_tokens and q_chunk_size >= q_tokens
    • skipping all chunking when the [email protected] matmul requires fewer bytes than a user-provided threshold
    opened by Birch-san 1
Releases(0.1.3)
  • 0.1.2(Mar 7, 2022)

    What's Changed

    This update fixes torch device handling issues in code. GPU and other kinds of tensors can be used safely.

    • Update utils.py by @yhgon in https://github.com/AminRezaei0x443/memory-efficient-attention/pull/5
    • Update attention_torch.py by @yhgon in https://github.com/AminRezaei0x443/memory-efficient-attention/pull/6

    New Contributors

    • @yhgon made their first contribution in https://github.com/AminRezaei0x443/memory-efficient-attention/pull/5

    Full Changelog: https://github.com/AminRezaei0x443/memory-efficient-attention/compare/0.1.1.0...0.1.2

    Source code(tar.gz)
    Source code(zip)
  • 0.1.1.0(Feb 3, 2022)

    Added mask, bias calculation functions for custom and memory efficient chunks computation. So now sublinear memory computation mask, bias are possible.

    Full Changelog: https://github.com/AminRezaei0x443/memory-efficient-attention/compare/0.1.1...0.1.1.0

    Source code(tar.gz)
    Source code(zip)
Owner
Amin Rezaei
Computer Science BSc, Neural Networks Enthusiast
Amin Rezaei
TensorFlow 101: Introduction to Deep Learning for Python Within TensorFlow

TensorFlow 101: Introduction to Deep Learning I have worked all my life in Machine Learning, and I've never seen one algorithm knock over its benchmar

Sefik Ilkin Serengil 896 Jan 04, 2023
Ludwig is a toolbox that allows to train and evaluate deep learning models without the need to write code.

Translated in 🇰🇷 Korean/ Ludwig is a toolbox that allows users to train and test deep learning models without the need to write code. It is built on

Ludwig 8.7k Dec 31, 2022
Code, Data and Demo for Paper: Controllable Generation from Pre-trained Language Models via Inverse Prompting

InversePrompting Paper: Controllable Generation from Pre-trained Language Models via Inverse Prompting Code: The code is provided in the "chinese_ip"

THUDM 101 Dec 16, 2022
a minimal terminal with python 😎😉

Meterm a terminal with python 😎 How to use Clone Project: $ git clone https://github.com/motahharm/meterm.git Run: in Terminal: meterm.exe Or pip ins

Motahhar.Mokfi 5 Jan 28, 2022
2021 CCF BDCI 全国信息检索挑战杯(CCIR-Cup)智能人机交互自然语言理解赛道第二名参赛解决方案

2021 CCF BDCI 全国信息检索挑战杯(CCIR-Cup) 智能人机交互自然语言理解赛道第二名解决方案 比赛网址: CCIR-Cup-智能人机交互自然语言理解 1.依赖环境: python==3.8 torch==1.7.1+cu110 numpy==1.19.2 transformers=

JinXiang 22 Oct 29, 2022
Repo for the Video Person Clustering dataset, and code for the associated paper

Video Person Clustering Repo for the Video Person Clustering dataset, and code for the associated paper. This reporsitory contains the Video Person Cl

Andrew Brown 47 Nov 02, 2022
Code for NeurIPS 2021 paper 'Spatio-Temporal Variational Gaussian Processes'

Spatio-Temporal Variational GPs This repository is the official implementation of the methods in the publication: O. Hamelijnck, W.J. Wilkinson, N.A.

AaltoML 26 Sep 16, 2022
Official code release for ICCV 2021 paper SNARF: Differentiable Forward Skinning for Animating Non-rigid Neural Implicit Shapes.

Official code release for ICCV 2021 paper SNARF: Differentiable Forward Skinning for Animating Non-rigid Neural Implicit Shapes.

235 Dec 26, 2022
CDTrans: Cross-domain Transformer for Unsupervised Domain Adaptation

[ICCV2021] TransReID: Transformer-based Object Re-Identification [pdf] The official repository for TransReID: Transformer-based Object Re-Identificati

DamoCV 569 Dec 30, 2022
Adaptive Pyramid Context Network for Semantic Segmentation (APCNet CVPR'2019)

Adaptive Pyramid Context Network for Semantic Segmentation (APCNet CVPR'2019) Introduction Official implementation of Adaptive Pyramid Context Network

21 Nov 09, 2022
Simple Python project using Opencv and datetime package to recognise faces and log attendance data in a csv file.

Attendance-System-based-on-Facial-recognition-Attendance-data-stored-in-csv-file- Simple Python project using Opencv and datetime package to recognise

3 Aug 09, 2022
Data Augmentation Using Keras and Python

Data-Augmentation-Using-Keras-and-Python Data augmentation is the process of increasing the number of training dataset. Keras library offers a simple

Happy N. Monday 3 Feb 15, 2022
ClevrTex: A Texture-Rich Benchmark for Unsupervised Multi-Object Segmentation

ClevrTex This repository contains dataset generation code for ClevrTex benchmark from paper: ClevrTex: A Texture-Rich Benchmark for Unsupervised Multi

Laurynas Karazija 26 Dec 21, 2022
Model-free Vehicle Tracking and State Estimation in Point Cloud Sequences

Model-free Vehicle Tracking and State Estimation in Point Cloud Sequences 1. Introduction This project is for paper Model-free Vehicle Tracking and St

TuSimple 92 Jan 03, 2023
ICCV2021: Code for 'Spatial Uncertainty-Aware Semi-Supervised Crowd Counting'

ICCV2021: Code for 'Spatial Uncertainty-Aware Semi-Supervised Crowd Counting'

Yanda Meng 14 May 13, 2022
计算机视觉中用到的注意力模块和其他即插即用模块PyTorch Implementation Collection of Attention Module and Plug&Play Module

PyTorch实现多种计算机视觉中网络设计中用到的Attention机制,还收集了一些即插即用模块。由于能力有限精力有限,可能很多模块并没有包括进来,有任何的建议或者改进,可以提交issue或者进行PR。

PJDong 599 Dec 23, 2022
PyTorch implementation for Stochastic Fine-grained Labeling of Multi-state Sign Glosses for Continuous Sign Language Recognition.

Stochastic CSLR This is the PyTorch implementation for the ECCV 2020 paper: Stochastic Fine-grained Labeling of Multi-state Sign Glosses for Continuou

Zhe Niu 28 Dec 19, 2022
Doge-Prediction - Coding Club prediction ig

Doge-Prediction Coding Club prediction ig Basically: Create an application that

1 Jan 10, 2022
Visual Question Answering in Pytorch

Visual Question Answering in pytorch /!\ New version of pytorch for VQA available here: https://github.com/Cadene/block.bootstrap.pytorch This repo wa

Remi 672 Jan 01, 2023
DNA sequence classification by Deep Neural Network

DNA sequence classification by Deep Neural Network: Project Overview worked on the DNA sequence classification problem where the input is the DNA sequ

Mohammed Jawwadul Islam Fida 0 Aug 02, 2022