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
[ICCV 2021] HRegNet: A Hierarchical Network for Large-scale Outdoor LiDAR Point Cloud Registration

HRegNet: A Hierarchical Network for Large-scale Outdoor LiDAR Point Cloud Registration Introduction The repository contains the source code and pre-tr

Intelligent Sensing, Perception and Computing Group 55 Dec 14, 2022
Code for BMVC2021 "MOS: A Low Latency and Lightweight Framework for Face Detection, Landmark Localization, and Head Pose Estimation"

MOS-Multi-Task-Face-Detect Introduction This repo is the official implementation of "MOS: A Low Latency and Lightweight Framework for Face Detection,

104 Dec 08, 2022
[TNNLS 2021] The official code for the paper "Learning Deep Context-Sensitive Decomposition for Low-Light Image Enhancement"

CSDNet-CSDGAN this is the code for the paper "Learning Deep Context-Sensitive Decomposition for Low-Light Image Enhancement" Environment Preparing pyt

Jiaao Zhang 17 Nov 05, 2022
Repository for the "Gotta Go Fast When Generating Data with Score-Based Models" paper

Gotta Go Fast When Generating Data with Score-Based Models This repo contains the official implementation for the paper Gotta Go Fast When Generating

Alexia Jolicoeur-Martineau 89 Nov 09, 2022
A series of Python scripts to access measurements from Fluke 28X meters. Fluke IR Remote Interface required.

Fluke289_data_access A series of Python scripts to access measurements from Fluke 28X meters. Fluke IR Remote Interface required. Created from informa

3 Dec 08, 2022
Publication describing 3 ML examples at NSLS-II and interfacing into Bluesky

Machine learning enabling high-throughput and remote operations at large-scale user facilities. Overview This repository contains the source code and

BNL 4 Sep 24, 2022
Code of paper "Compositionally Generalizable 3D Structure Prediction"

Compositionally Generalizable 3D Structure Prediction In this work, We bring in the concept of compositional generalizability and factorizes the 3D sh

Songfang Han 30 Dec 17, 2022
Graph Convolutional Neural Networks with Data-driven Graph Filter (GCNN-DDGF)

Graph Convolutional Gated Recurrent Neural Network (GCGRNN) Improved from Graph Convolutional Neural Networks with Data-driven Graph Filter (GCNN-DDGF

Lei Lin 21 Dec 18, 2022
Implementation of our recent paper, WOOD: Wasserstein-based Out-of-Distribution Detection.

WOOD Implementation of our recent paper, WOOD: Wasserstein-based Out-of-Distribution Detection. Abstract The training and test data for deep-neural-ne

8 Dec 24, 2022
A tensorflow/keras implementation of StyleGAN to generate images of new Pokemon.

PokeGAN A tensorflow/keras implementation of StyleGAN to generate images of new Pokemon. Dataset The model has been trained on dataset that includes 8

19 Jul 26, 2022
The pure and clear PyTorch Distributed Training Framework.

The pure and clear PyTorch Distributed Training Framework. Introduction Requirements and Usage Dependency Dataset Basic Usage Slurm Cluster Usage Base

WILL LEE 208 Dec 20, 2022
A collection of resources and papers on Diffusion Models, a darkhorse in the field of Generative Models

This repository contains a collection of resources and papers on Diffusion Models and Score-based Models. If there are any missing valuable resources

5.1k Jan 08, 2023
Container : Context Aggregation Network

Container : Context Aggregation Network If you use this code for a paper please cite: @article{gao2021container, title={Container: Context Aggregati

AI2 47 Dec 16, 2022
MetaDrive: Composing Diverse Scenarios for Generalizable Reinforcement Learning

MetaDrive: Composing Diverse Driving Scenarios for Generalizable RL [ Documentation | Demo Video ] MetaDrive is a driving simulator with the following

DeciForce: Crossroads of Machine Perception and Autonomy 276 Jan 04, 2023
A PyTorch implementation of the Relational Graph Convolutional Network (RGCN).

Torch-RGCN Torch-RGCN is a PyTorch implementation of the RGCN, originally proposed by Schlichtkrull et al. in Modeling Relational Data with Graph Conv

Thiviyan Singam 66 Nov 30, 2022
CLIP (Contrastive Language–Image Pre-training) for Italian

Italian CLIP CLIP (Radford et al., 2021) is a multimodal model that can learn to represent images and text jointly in the same space. In this project,

Italian CLIP 114 Dec 29, 2022
Using machine learning to predict and analyze high and low reader engagement for New York Times articles posted to Facebook.

How The New York Times can increase Engagement on Facebook Using machine learning to understand characteristics of news content that garners "high" Fa

Jessica Miles 0 Sep 16, 2021
Author's PyTorch implementation of TD3 for OpenAI gym tasks

Addressing Function Approximation Error in Actor-Critic Methods PyTorch implementation of Twin Delayed Deep Deterministic Policy Gradients (TD3). If y

Scott Fujimoto 1.3k Dec 25, 2022
Official pytorch implementation of the IrwGAN for unaligned image-to-image translation

IrwGAN (ICCV2021) Unaligned Image-to-Image Translation by Learning to Reweight [Update] 12/15/2021 All dataset are released, trained models and genera

37 Nov 09, 2022
Deep learning operations reinvented (for pytorch, tensorflow, jax and others)

This video in better quality. einops Flexible and powerful tensor operations for readable and reliable code. Supports numpy, pytorch, tensorflow, and

Alex Rogozhnikov 6.2k Jan 01, 2023