PyTea: PyTorch Tensor shape error analyzer

Overview

PyTea: PyTorch Tensor Shape Error Analyzer

paper project page

Requirements

  • node.js >= 12.x
  • python >= 3.8
    • z3-solver >= 4.8

How to install and use

# install node.js
sudo apt-get install nodejs

# install python z3-solver
pip install z3-solver

# download pytea
wget https://github.com/ropas/pytea/releases/download/v0.1.0/pytea.zip
unzip pytea.zip

# run pytea
python bin/pytea.py path/to/source.py

# run example file
python bin/pytea.py packages/pytea/pytest/basics/scratch.py

How to build

# install dependencies
npm run install:all
pip install z3-solver

# build
npm run build

Documentations

Brief explanation of the analysis result

PyTea is composed of two analyzers.

  • Online analysis: node.js (TypeScript / JavaScript)
    • Find numeric range-based shape mismatch and misuse of API argument. If PyTea has found any error while analyzing the code, it will stop at that position and inform the errors and violated constraints to the user.
  • Offline analysis: Z3 / Python
    • The generated constraints are passed to Z3Py. Z3 will solve the constraint sets of each path and print the first violated constraint (if it exists).

The result of the Online analyzer is divided into three classes:

  • potential success path: the analyzer does not found shape mismatch until now, but the final constraint set can be violated if Z3 analyzes it on closer inspection.
  • potential unreachable path: the analyzer found a shape mismatch or API misuses, but there remain path constraints. In short, path constraint is an unresolved branch condition; that means the stopped path might be unreachable if remaining path constraints have a contradiction. Those cases will be distinguished from Offline analysis.
  • immediate failed path: the analyzer has found an error, stops its analysis immediately.

CAVEAT: If the code contains PyTorch or other third-party APIs that we have not implemented, it will raise false alarms. Nevertheless, we also record each unimplemented API call. See LOGS section from the result and search which unimplemented API call is performed.

The final result of the Offline analysis is divided into several cases.

  • Valid path: SMT solver has not found any error. Every constraint will always be fulfilled.
  • Invalid path: SMT solver found a condition that can violate some constraints. Notice that this does not mean the code will always crash, but it found an extreme case that crashes some executions.
  • Undecidable path: SMT solver has met unsolvable constraints, then timeouts. Some non-linear formulae can be classified into this case.
  • Unreachable path: Hard and Path constraints contain contradicting constraints; this path will not be realized from the beginning.

Result examples

  • Error found by Online analysis

test1

  • Error found by Offline analysis

test2

License

MIT License

This project is based on Pyright, also MIT License

Comments
  • LSTM/GRU input_size tensor shape errors

    LSTM/GRU input_size tensor shape errors

    Hi, I have met a problem while detecting LSTM tensor shape errors. The testing file below is runnable and pytea returns correctly.

    import torch
    import torch.nn as nn
    import torch.nn.functional as F
    
    
    class Model(nn.Module):
    
        def __init__(self):
            super().__init__()
            self.cnn = nn.Conv2d(3, 1, 3, 1, 1)
            self.rnn = nn.LSTM(32, 64, 1, batch_first=True)
            self.pool = nn.MaxPool2d(2, 2)
            self.fc = nn.Linear(64, 16)
    
        def forward(self, x):
            x = self.pool(F.relu(self.cnn(x)))
            x = x.view(-1, 32, 32)
            x, _ = self.rnn(x)
            x = x[:, -1, :].squeeze(1)
            x = F.relu(self.fc(x))
            x = F.softmax(x, dim=-1)
            return x
    
    
    if __name__ == "__main__":
        net = Model()
        x = torch.randn(2, 3, 64, 64)
        y = net(x)
        target = torch.argmax(torch.randn(2, 16), dim=-1)
        loss = F.cross_entropy(y, target.long())
        loss.backward()
        print(y.size())
    

    However, If I change self.rnn = nn.LSTM(32, 64, 1, batch_first=True) into self.rnn = nn.LSTM(64, 64, 1, batch_first=True), torch will report a RuntimeError: Expected 64, got 32. pytea didn't return any CONSTRAINTS information, as it supposed to.

    Then I tried to more LSTM input_size shape errors, all failed. Same situation with GRU. I think it is a bug, because I can detect Conv2d, Linear error successfully.

    opened by MCplayerFromPRC 4
  • Inconsistent with document description

    Inconsistent with document description

    Hello, I have encountered the following problems:

    First question: The content of my source file is:

       import torch
       import torch.nn as nn
    
      class Net(nn.Module):
          def __init__(self):
              super(Net, self).__init__()
              self.layers = nn.Sequential(
                  nn.Linear(28 * 28, 120),
                  nn.ReLU(),
                  nn.Linear(80, 10))
      
          def a(self):
              pass
    
        if __name__ == "__main__":
            n = Net()
    

    But when I execute the command, I get the following results:

    image

    There should be a problem with defining shape in this model.

    Second question: I used it https://github.com/pytorch/examples/blob/master/mnist/main.py , but the command is stuck and no result is returned. As follows:

    image
    opened by dejavu6 2
  • Ternary expression (A if B else C) bug report

    Ternary expression (A if B else C) bug report

    아래와 같은 코드 실행에서 문제가 발생한다는 것을 깨달았습니다.

    x = 0 if 0 <= 1 else 1
    
    # runtime output
    REDUCED HEAP: (size: 250)
      x => 1
    

    파이썬의 삼항연산자가 x = (((0 <= 1) and 0) or 1)로 파싱됩니다. Logical statement가 True, true-value가 0일 때 발생하는 오류인 것으로 보입니다.

    당장 벤치마크 코드에서 나타나는 문제는 아닙니다. Pylib builtin 구현에서 발생한 문제이므로, 다른 방식으로 구현함으로써 일단은 피해갈 수 있을 것 같습니다.

    감사합니다.

    opened by lego0901 1
  • Develop sehoon

    Develop sehoon

    UT-1~6 코드 분석에 필요한 torch API 구현 완료.

    UT-2: epoch을 1로 수정하지 않으면 timeout이됨. UT-3: Python 빌트인 함수 iter, next의 구현은 우선 넘어갔음. UT-6: buggy 코드에서 target이 free variable인데, 이를 처리해 주지 않고 분석을 실행하면 아무것도 출력하지 않는 버그가 있음.

    위 특이사항을 적절히 처리해주고 분석을 실행하면 1~6 모두 buggy 코드는 invalid, fix 코드는 valid 결과를 냄.

    opened by Sehun0819 0
  • path constraint check

    path constraint check

    분석중 한 패스에서 (텐서 모양 오류 등의) 에러를 만나면 해당 패스는 처리됨. 문제는 분기 조건문에 의해 실제로는 진행되지 않는 패스여도 로 처리가 되는 것.

    따라서 분석중 에러를 만났을 때 그 패스가 path constraint를 갖고 있으면 로 처리하여 z3단에 넘기게 수정하였음.

    TODO: z3단에 넘기기 전에 path constraint를 계산하여 Valid면 , Unsat이면 로 처리하기(else )

    opened by Sehun0819 0
  • Bump node-notifier from 8.0.0 to 8.0.1 in /packages/pyright-internal

    Bump node-notifier from 8.0.0 to 8.0.1 in /packages/pyright-internal

    Bumps node-notifier from 8.0.0 to 8.0.1.

    Changelog

    Sourced from node-notifier's changelog.

    v8.0.1

    • fixes possible injection issue for notify-send
    Commits

    Dependabot compatibility score

    Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.


    Dependabot commands and options

    You can trigger Dependabot actions by commenting on this PR:

    • @dependabot rebase will rebase this PR
    • @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
    • @dependabot merge will merge this PR after your CI passes on it
    • @dependabot squash and merge will squash and merge this PR after your CI passes on it
    • @dependabot cancel merge will cancel a previously requested merge and block automerging
    • @dependabot reopen will reopen this PR if it is closed
    • @dependabot close will close this PR and stop Dependabot recreating it. You can achieve the same result by closing it manually
    • @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
    • @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
    • @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)
    • @dependabot use these labels will set the current labels as the default for future PRs for this repo and language
    • @dependabot use these reviewers will set the current reviewers as the default for future PRs for this repo and language
    • @dependabot use these assignees will set the current assignees as the default for future PRs for this repo and language
    • @dependabot use this milestone will set the current milestone as the default for future PRs for this repo and language

    You can disable automated security fix PRs for this repo from the Security Alerts page.

    dependencies 
    opened by dependabot[bot] 0
Releases(v0.1.0)
Owner
ROPAS Lab.
ROPAS Lab. @ Seoul National University
ROPAS Lab.
Implementation of the paper Recurrent Glimpse-based Decoder for Detection with Transformer.

REGO-Deformable DETR By Zhe Chen, Jing Zhang, and Dacheng Tao. This repository is the implementation of the paper Recurrent Glimpse-based Decoder for

Zhe Chen 33 Nov 30, 2022
Code for paper "ASAP-Net: Attention and Structure Aware Point Cloud Sequence Segmentation"

ASAP-Net This project implements ASAP-Net of paper ASAP-Net: Attention and Structure Aware Point Cloud Sequence Segmentation (BMVC2020). Overview We i

Hanwen Cao 26 Aug 25, 2022
Do you like Quick, Draw? Well what if you could train/predict doodles drawn inside Streamlit? Also draws lines, circles and boxes over background images for annotation.

Streamlit - Drawable Canvas Streamlit component which provides a sketching canvas using Fabric.js. Features Draw freely, lines, circles, boxes and pol

Fanilo Andrianasolo 325 Dec 28, 2022
Geometric Sensitivity Decomposition

Geometric Sensitivity Decomposition This repo is the official implementation of A Geometric Perspective towards Neural Calibration via Sensitivity Dec

16 Dec 26, 2022
DECA: Detailed Expression Capture and Animation (SIGGRAPH 2021)

DECA: Detailed Expression Capture and Animation (SIGGRAPH2021) input image, aligned reconstruction, animation with various poses & expressions This is

Yao Feng 1.5k Jan 02, 2023
Code for the paper "Controllable Video Captioning with an Exemplar Sentence"

SMCG Code for the paper "Controllable Video Captioning with an Exemplar Sentence" Introduction We investigate a novel and challenging task, namely con

10 Dec 04, 2022
Yet Another Reinforcement Learning Tutorial

This repo contains self-contained RL implementations

Sungjoon 65 Dec 10, 2022
Differentiable Simulation of Soft Multi-body Systems

Differentiable Simulation of Soft Multi-body Systems Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin [Paper] [Code] Updates The C++ backend s

YilingQiao 26 Dec 23, 2022
Code repo for "Cross-Scale Internal Graph Neural Network for Image Super-Resolution" (NeurIPS'20)

IGNN Code repo for "Cross-Scale Internal Graph Neural Network for Image Super-Resolution" [paper] [supp] Prepare datasets 1 Download training dataset

Shangchen Zhou 278 Jan 03, 2023
This is a collection of our NAS and Vision Transformer work.

AutoML - Neural Architecture Search This is a collection of our AutoML-NAS work iRPE (NEW): Rethinking and Improving Relative Position Encoding for Vi

Microsoft 832 Jan 08, 2023
MiniSom is a minimalistic implementation of the Self Organizing Maps

MiniSom Self Organizing Maps MiniSom is a minimalistic and Numpy based implementation of the Self Organizing Maps (SOM). SOM is a type of Artificial N

Giuseppe Vettigli 1.2k Jan 03, 2023
Scikit-learn compatible estimation of general graphical models

skggm : Gaussian graphical models using the scikit-learn API In the last decade, learning networks that encode conditional independence relationships

213 Jan 02, 2023
alfred-py: A deep learning utility library for **human**

Alfred Alfred is command line tool for deep-learning usage. if you want split an video into image frames or combine frames into a single video, then a

JinTian 800 Jan 03, 2023
PSANet: Point-wise Spatial Attention Network for Scene Parsing, ECCV2018.

PSANet: Point-wise Spatial Attention Network for Scene Parsing (in construction) by Hengshuang Zhao*, Yi Zhang*, Shu Liu, Jianping Shi, Chen Change Lo

Hengshuang Zhao 217 Oct 30, 2022
A TensorFlow 2.x implementation of Masked Autoencoders Are Scalable Vision Learners

Masked Autoencoders Are Scalable Vision Learners A TensorFlow implementation of Masked Autoencoders Are Scalable Vision Learners [1]. Our implementati

Aritra Roy Gosthipaty 59 Dec 10, 2022
BDDM: Bilateral Denoising Diffusion Models for Fast and High-Quality Speech Synthesis

Bilateral Denoising Diffusion Models (BDDMs) This is the official PyTorch implementation of the following paper: BDDM: BILATERAL DENOISING DIFFUSION M

172 Dec 23, 2022
Reading Group @mila-iqia on Computational Optimal Transport for Machine Learning Applications

Computational Optimal Transport for Machine Learning Reading Group Over the last few years, optimal transport (OT) has quickly become a central topic

Ali Harakeh 11 Aug 26, 2022
"Moshpit SGD: Communication-Efficient Decentralized Training on Heterogeneous Unreliable Devices", official implementation

Moshpit SGD: Communication-Efficient Decentralized Training on Heterogeneous Unreliable Devices This repository contains the official PyTorch implemen

Yandex Research 21 Oct 18, 2022
Implementation of the Point Transformer layer, in Pytorch

Point Transformer - Pytorch Implementation of the Point Transformer self-attention layer, in Pytorch. The simple circuit above seemed to have allowed

Phil Wang 501 Jan 03, 2023
Measures input lag without dedicated hardware, performing motion detection on recorded or live video

What is InputLagTimer? This tool can measure input lag by analyzing a video where both the game controller and the game screen can be seen on a webcam

Bruno Gonzalez 4 Aug 18, 2022