QuanTaichi evaluation suite

Overview

QuanTaichi: A Compiler for Quantized Simulations (SIGGRAPH 2021)

Yuanming Hu, Jiafeng Liu, Xuanda Yang, Mingkuan Xu, Ye Kuang, Weiwei Xu, Qiang Dai, William T. Freeman, Fredo Durand

[Paper] [Video]

The QuanTaichi framework is now officially part of Taichi. This repo only contains examples.

Simulate more with less memory, using a quantization compiler.

High-resolution simulations can deliver great visual quality, but they are often limited by available memory. We present a compiler for physical simulation that can achieve both high performance and significantly reduced memory costs, by enabling flexible and aggressive quantization.

To achieve that, we implemented an extension of the type system in Taichi. Now, programmers can define custom data types using the following code:

i8 = ti.quant.int(bits=8, signed=True)
fixed12 = ti.quant.fixed(frac=12, signed=False, range=3.0)
cft16 = ti.quant.float(exp=5, frac=11, signed=True)

The compiler will automatically encode/decode numerical data to achieve an improved memory efficiency (storage & bandwidth). Since custom data types are not natively supported by hardware, we propose two useful types of bit adapters: Bit structs and Bit arrays to pack thses types into hardware supported types with bit width 8, 16, 32, 64. For example, The following code declears 2 fields with custom types, and materialized them into two 2D 4 x 2 arrays with Bit structs:

u4 = ti.quant.int(bits=4, signed=False)
i12 = ti.quant.int(bits=12, signed=True)
p = ti.field(dtype=u4)
q = ti.field(dtype=i12)
ti.root.dense(ti.ij, (4, 2)).bit_struct(num_bits=16).place(p, q)

The p and q fields are laid in an array of structure (AOS) order in memory. Note the containing bit struct of a (p[i, j], q[i, j]) tuple is 16-bit wide. For more details of the usage of our quantization type system, please refer to our paper or see the examples in this repo.

Under proper quantization, we achieve 8× higher memory efficiency on each Game of Life cell, 1.57× on each Eulerian fluid simulation voxel, and 1.7× on each material point method particle. To the best of our knowledge, this is the first time these high-resolution simulations can run on a single GPU. Our system achieves resolution, performance, accuracy, and visual quality simultaneously.

How to run

Install the latest Taichi first.

Install the latest Taichi by:

python3 -m pip install —U taichi

Game of Life (GoL)

gol_pic

To reproduce the GOL galaxy:

cd gol && python3 galaxy.py -a [cpu/cuda] -o output

We suggest you run the script using GPU (--arch cuda). Because to better observe the evolution of metapixels, we set the steps per frame to be 32768 which will take quite a while on CPUs.

To reproduce the super large scale GoL:

  1. Download the pattern quant_sim_meta.rle from our Google Drive and place it in the same folder with quant_sim.py

  2. Run the code

python3 quant_sim.py -a [cpu/cuda] -o output

For more details, please refer to this documentation.

MLS-MPM

mpm-pic

To test our system on hybrid Lagrangian-Eulerian methods where both particles and grids are used, we implemented the Moving Least Squares Material Point Method with G2P2G transfer.

To reproduce, please see the output of the following command:

cd mls-mpm
python3 -m demo.demo_quantized_simulation_letters --help

You can add -s flag for a quick visualization and you may need to wait for 30 frames to see letters falling down.

More details are in this documentation.

Eulerian Fluid

smoke_simulation

We developed a sparse-grid-based advection-reflection fluid solver to evaluate our system on grid-based physical simulators.

To reproduce the large scale smoke simulation demo, please first change the directory into eulerain_fluid, and run:

python3 run.py --demo [0/1] -o outputs

Set the arg of demo to 0 for the bunny demo and 1 for the flow demo. -o outputs means the set the output folder to outputs.

For more comparisons of this quantized fluid simulation, please refer to the documentation of this demo.

Microbenchmarks

To reproduce the experiments of microbenchmarks, please run

cd microbenchmarks
chmod +x run_microbenchmarks.sh
./run_microbenchmarks.sh

Please refer to this Readme to get more details.

Bibtex

@article{hu2021quantaichi,
  title={QuanTaichi: A Compiler for Quantized Simulations},
  author={Hu, Yuanming and Liu, Jiafeng and Yang, Xuanda and Xu, Mingkuan and Kuang, Ye and Xu, Weiwei and Dai, Qiang and Freeman, William T. and Durand, Frédo},
  journal={ACM Transactions on Graphics (TOG)},
  volume={40},
  number={4},
  year={2021},
  publisher={ACM}
}
Owner
Taichi Developers
Taichi Developers
An Open-Source Toolkit for Prompt-Learning.

An Open-Source Framework for Prompt-learning. Overview • Installation • How To Use • Docs • Paper • Citation • What's New? Nov 2021: Now we have relea

THUNLP 2.3k Jan 07, 2023
The official code for paper "R2D2: Recursive Transformer based on Differentiable Tree for Interpretable Hierarchical Language Modeling".

R2D2 This is the official code for paper titled "R2D2: Recursive Transformer based on Differentiable Tree for Interpretable Hierarchical Language Mode

Alipay 49 Dec 17, 2022
Optimized Gillespie algorithm for simulating Stochastic sPAtial models of Cancer Evolution (OG-SPACE)

OG-SPACE Introduction Optimized Gillespie algorithm for simulating Stochastic sPAtial models of Cancer Evolution (OG-SPACE) is a computational framewo

Data and Computational Biology Group UNIMIB (was BI*oinformatics MI*lan B*icocca) 0 Nov 17, 2021
PyTorch implementation for the paper Visual Representation Learning with Self-Supervised Attention for Low-Label High-Data Regime

Visual Representation Learning with Self-Supervised Attention for Low-Label High-Data Regime Created by Prarthana Bhattacharyya. Disclaimer: This is n

Prarthana Bhattacharyya 5 Nov 08, 2022
An adaptive hierarchical energy management strategy for hybrid electric vehicles

An adaptive hierarchical energy management strategy This project contains the source code of an adaptive hierarchical EMS combining heuristic equivale

19 Dec 13, 2022
3D-aware GANs based on NeRF (arXiv).

CIPS-3D This repository will contain the code of the paper, CIPS-3D: A 3D-Aware Generator of GANs Based on Conditionally-Independent Pixel Synthesis.

Peterou 563 Dec 31, 2022
A Convolutional Transformer for Keyword Spotting

☢️ Audiomer ☢️ Audiomer: A Convolutional Transformer for Keyword Spotting [ arXiv ] [ Previous SOTA ] [ Model Architecture ] Results on SpeechCommands

49 Jan 27, 2022
SE3 Pose Interp - Interpolate camera pose or trajectory in SE3, pose interpolation, trajectory interpolation

SE3 Pose Interpolation Pose estimated from SLAM system are always discrete, and

Ran Cheng 4 Dec 15, 2022
Inverse Optimal Control Adapted to the Noise Characteristics of the Human Sensorimotor System

Inverse Optimal Control Adapted to the Noise Characteristics of the Human Sensorimotor System This repository contains code for the paper Schultheis,

2 Oct 28, 2022
RRL: Resnet as representation for Reinforcement Learning

Resnet as representation for Reinforcement Learning (RRL) is a simple yet effective approach for training behaviors directly from visual inputs. We demonstrate that features learned by standard image

Meta Research 21 Dec 07, 2022
(CVPR 2021) Lifting 2D StyleGAN for 3D-Aware Face Generation

Lifting 2D StyleGAN for 3D-Aware Face Generation Official implementation of paper "Lifting 2D StyleGAN for 3D-Aware Face Generation". Requirements You

Yichun Shi 66 Nov 29, 2022
Random Erasing Data Augmentation. Experiments on CIFAR10, CIFAR100 and Fashion-MNIST

Random Erasing Data Augmentation =============================================================== black white random This code has the source code for

Zhun Zhong 654 Dec 26, 2022
Contrastive Fact Verification

VitaminC This repository contains the dataset and models for the NAACL 2021 paper: Get Your Vitamin C! Robust Fact Verification with Contrastive Evide

47 Dec 19, 2022
Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS2021)

Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS2021) Overview Prerequisites Linux Pytho

Shaojie Li 34 Mar 31, 2022
Graduation Project

Gesture-Detection-and-Depth-Estimation This is my graduation project. (1) In this project, I use the YOLOv3 object detection model to detect gesture i

ChaosAT 1 Nov 23, 2021
Implementation of Rotary Embeddings, from the Roformer paper, in Pytorch

Rotary Embeddings - Pytorch A standalone library for adding rotary embeddings to transformers in Pytorch, following its success as relative positional

Phil Wang 110 Dec 30, 2022
Machine Learning Model deployment for Container (TensorFlow Serving)

try_tf_serving ├───dataset │ ├───testing │ │ ├───paper │ │ ├───rock │ │ └───scissors │ └───training │ ├───paper │ ├───rock

Azhar Rizki Zulma 5 Jan 07, 2022
the official implementation of the paper "Isometric Multi-Shape Matching" (CVPR 2021)

Isometric Multi-Shape Matching (IsoMuSh) Paper-CVF | Paper-arXiv | Video | Code Citation If you find our work useful in your research, please consider

Maolin Gao 9 Jul 17, 2022
Code for "OctField: Hierarchical Implicit Functions for 3D Modeling (NeurIPS 2021)"

OctField(Jittor): Hierarchical Implicit Functions for 3D Modeling Introduction This repository is code release for OctField: Hierarchical Implicit Fun

55 Dec 08, 2022
Makes patches from huge resolution .svs slide files using openslide

openslide_patcher Makes patches from huge resolution .svs slide files using openslide Example collage I made from outputs:

2 Dec 23, 2021