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
[CVPR 2022 Oral] Crafting Better Contrastive Views for Siamese Representation Learning

Crafting Better Contrastive Views for Siamese Representation Learning (CVPR 2022 Oral) 2022-03-29: The paper was selected as a CVPR 2022 Oral paper! 2

249 Dec 28, 2022
Applying CLIP to Point Cloud Recognition.

PointCLIP: Point Cloud Understanding by CLIP This repository is an official implementation of the paper 'PointCLIP: Point Cloud Understanding by CLIP'

Renrui Zhang 175 Dec 24, 2022
With this package, you can generate mixed-integer linear programming (MIP) models of trained artificial neural networks (ANNs) using the rectified linear unit (ReLU) activation function

With this package, you can generate mixed-integer linear programming (MIP) models of trained artificial neural networks (ANNs) using the rectified linear unit (ReLU) activation function. At the momen

ChemEngAI 40 Dec 27, 2022
Deep motion generator collections

GenMotion GenMotion (/gen’motion/) is a Python library for making skeletal animations. It enables easy dataset loading and experiment sharing for synt

23 May 24, 2022
Understanding and Overcoming the Challenges of Efficient Transformer Quantization

Transformer Quantization This repository contains the implementation and experiments for the paper presented in Yelysei Bondarenko1, Markus Nagel1, Ti

83 Dec 30, 2022
Discord-Protect is a simple discord bot allowing you to have some security on your discord server by ordering a captcha to the user who joins your server.

Discord-Protect Discord-Protect is a simple discord bot allowing you to have some security on your discord server by ordering a captcha to the user wh

Tir Omar 2 Oct 28, 2021
MakeItTalk: Speaker-Aware Talking-Head Animation

MakeItTalk: Speaker-Aware Talking-Head Animation This is the code repository implementing the paper: MakeItTalk: Speaker-Aware Talking-Head Animation

Adobe Research 285 Jan 08, 2023
Python package to generate image embeddings with CLIP without PyTorch/TensorFlow

imgbeddings A Python package to generate embedding vectors from images, using OpenAI's robust CLIP model via Hugging Face transformers. These image em

Max Woolf 81 Jan 04, 2023
Certified Patch Robustness via Smoothed Vision Transformers

Certified Patch Robustness via Smoothed Vision Transformers This repository contains the code for replicating the results of our paper: Certified Patc

Madry Lab 35 Dec 14, 2022
Repositorio de los Laboratorios de Análisis Numérico / Análisis Numérico I de FAMAF, UNC.

Repositorio de los Laboratorios de Análisis Numérico / Análisis Numérico I de FAMAF, UNC. Para los Laboratorios de la materia, vamos a utilizar el len

Luis Biedma 18 Dec 12, 2022
Code release to accompany paper "Geometry-Aware Gradient Algorithms for Neural Architecture Search."

Geometry-Aware Gradient Algorithms for Neural Architecture Search This repository contains the code required to run the experiments for the DARTS sear

18 May 27, 2022
PyTorch implementation of TSception V2 using DEAP dataset

TSception This is the PyTorch implementation of TSception V2 using DEAP dataset in our paper: Yi Ding, Neethu Robinson, Su Zhang, Qiuhao Zeng, Cuntai

Yi Ding 27 Dec 15, 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
Self-Regulated Learning for Egocentric Video Activity Anticipation

Self-Regulated Learning for Egocentric Video Activity Anticipation Introduction This is a Pytorch implementation of the model described in our paper:

qzhb 13 Sep 23, 2022
[ICLR 2022] Contact Points Discovery for Soft-Body Manipulations with Differentiable Physics

CPDeform Code and data for paper Contact Points Discovery for Soft-Body Manipulations with Differentiable Physics at ICLR 2022 (Spotlight). @InProceed

(Lester) Sizhe Li 29 Nov 29, 2022
Code for Blind Image Decomposition (BID) and Blind Image Decomposition network (BIDeN).

arXiv, porject page, paper Blind Image Decomposition (BID) Blind Image Decomposition is a novel task. The task requires separating a superimposed imag

64 Dec 20, 2022
PyTorch Code for NeurIPS 2021 paper Anti-Backdoor Learning: Training Clean Models on Poisoned Data.

Anti-Backdoor Learning PyTorch Code for NeurIPS 2021 paper Anti-Backdoor Learning: Training Clean Models on Poisoned Data. Check the unlearning effect

Yige-Li 51 Dec 07, 2022
Pytorch implementation of forward and inverse Haar Wavelets 2D

Pytorch implementation of forward and inverse Haar Wavelets 2D

Sergei Belousov 9 Oct 30, 2022
details on efforts to dump the Watermelon Games Paprium cart

Reminder, if you like these repos, fork them so they don't disappear https://github.com/ArcadeHustle/WatermelonPapriumDump/fork Big thanks to Fonzie f

Hustle Arcade 29 Dec 11, 2022
Semi-Supervised Graph Prototypical Networks for Hyperspectral Image Classification, IGARSS, 2021.

Semi-Supervised Graph Prototypical Networks for Hyperspectral Image Classification, IGARSS, 2021. Bobo Xi, Jiaojiao Li, Yunsong Li and Qian Du. Code f

Bobo Xi 7 Nov 03, 2022