# barasamit/quantattack architecture

> Architecture diagram of the GitHub repository barasamit/quantattack, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2025-05-23). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/barasamit/quantattack
- Repository: https://github.com/barasamit/quantattack

## Overview

##### Project type and purpose
- This is an open-source research tool (QuantAttack) implemented in PyTorch, designed to generate and evaluate adversarial attacks on dynamically quantized vision transformer models.
- Its primary use-cases are:
- Running different adversarial attack algorithms (one-to-one, many-to-many, universal)
- Measuring the impact of these attacks on inference time, memory, and energy
- Configuring experiments via YAML/Python config files

##### File structure
Top-level scripts (entry points):
- many_to_many_attack.py, one_to_one_attack.py, universal_attack.py, single_eval_only.py, attack.py
Core modules:
- attacks/ – implementations of attack algorithms (e.g., projected gradient descent)
- losses.py & utils/losses_utils.py – definitions and helpers for custom loss functions
- measures/ – evaluation scripts (e.g., clean_vs_quant.py)
- utils/ – general utilities subdivided into:
– data_utils.py (dataset loading & preprocessing)
– model_utils.py + load_RepQ_ViT.py + load_ptq4vit.py (model loading, dynamic quantization wrappers)
– attack_utils.py, defenses.py, general.py, init_collect_arrays.py, map_imagnet.py
- configs/ – YAML and Python files defining datasets, models, attacks, losses
- notebooks/ – Jupyter notebooks for post-processing or combining CSV outputs

##### README for additional insights
- The README highlights dynamic quantization vulnerabilities in vision transformers and describes experiment setup: config-driven, CLI-based.
- Dependencies: PyTorch, transformers, accelerate, bitsandbytes, datasets, fvcore, pandas.

##### Elements
- Blue for configuration & orchestration (Config Manager, CLI Controllers)
- Green for data & model (Data Pipeline, Model Loader)
- Red for computation cores (Attack Engine, Loss Modules)
- Yellow for evaluation (Measurement & Notebooks)

## Diagram (Mermaid)

```mermaid
flowchart TD
    %% CLI Controllers
    subgraph "CLI Controllers"
        direction TB
        ManyToMany["many_to_many_attack.py"] 
        OneToOne["one_to_one_attack.py"]
        Universal["universal_attack.py"]
        CLI_SingleEval["single_eval_only.py"]
    end

    %% Configuration Manager
    subgraph "Configuration Manager"
        direction TB
        AttackConfig["attack_config.yaml"]
        AttacksConfig["attacks_config.py"]
        DatasetClassConfig["dataset_class_config.yaml"]
        DatasetConfig["dataset_config.yaml"]
        LossesConfig["losses_config.yaml"]
        ModelConfig["model_config.yaml"]
    end

    %% Data Pipeline
    subgraph "Data Pipeline"
        direction TB
        DataUtils["data_utils.py"]
        MapImageNet["map_imagnet.py"]
        ImageNetMap["image_net_map.json"]
    end

    %% Model Loader & Quantizer
    subgraph "Model Loader & Quantizer"
        direction TB
        LoadRepQ["load_RepQ_ViT.py"]
        LoadPTQ["load_ptq4vit.py"]
        ModelUtils["model_utils.py"]
    end

    %% Attack Engine
    subgraph "Attack Engine"
        direction TB
        BaseAttack["attack.py"]
        PGDAttack["projected_gradient_descent.py"]
        AttackUtils["attack_utils.py"]
    end

    %% Loss & Defense Modules
    subgraph "Loss & Defense Modules"
        direction TB
        Losses["losses.py"]
        LossUtils["losses_utils.py"]
        Defenses["defenses.py"]
    end

    %% Measurement & Evaluation
    subgraph "Measurement & Evaluation"
        direction TB
        CleanVsQuant["clean_vs_quant.py"]
        EvalSingle["single_eval_only.py"]
        CombineCSV["combine_csvs.ipynb"]
    end

    %% Relationships
    ManyToMany -->|orchestrates| AttackConfig
    OneToOne -->|orchestrates| AttackConfig
    Universal -->|orchestrates| AttackConfig
    CLI_SingleEval -->|orchestrates| AttackConfig

    AttackConfig -->|configures| DataUtils
    AttackConfig -->|configures| LoadRepQ
    AttackConfig -->|configures| BaseAttack
    AttackConfig -->|configures| Losses

    DataUtils -->|feeds data to| LoadRepQ
    DataUtils -->|feeds data to| BaseAttack

    MapImageNet --> DataUtils
    ImageNetMap --> DataUtils

    LoadRepQ --> BaseAttack
    LoadPTQ --> BaseAttack
    ModelUtils --> LoadRepQ
    ModelUtils --> LoadPTQ

    BaseAttack -->|uses| Losses
    PGDAttack --> BaseAttack
    AttackUtils --> BaseAttack

    Losses -->|helper| LossUtils
    Losses -->|helper| Defenses

    BaseAttack -->|produces adversarial examples| CleanVsQuant
    EvalSingle --> CleanVsQuant
    CleanVsQuant --> CombineCSV

    %% Click Events
    click many_to_many_attack.py "https://github.com/barasamit/quantattack/blob/main/many_to_many_attack.py"
    click one_to_one_attack.py "https://github.com/barasamit/quantattack/blob/main/one_to_one_attack.py"
    click universal_attack.py "https://github.com/barasamit/quantattack/blob/main/universal_attack.py"
    click single_eval_only.py "https://github.com/barasamit/quantattack/blob/main/single_eval_only.py"

    click attack_config.yaml "https://github.com/barasamit/quantattack/blob/main/configs/attack_config.yaml"
    click attacks_config.py "https://github.com/barasamit/quantattack/blob/main/configs/attacks_config.py"
    click dataset_class_config.yaml "https://github.com/barasamit/quantattack/blob/main/configs/dataset_class_config.yaml"
    click dataset_config.yaml "https://github.com/barasamit/quantattack/blob/main/configs/dataset_config.yaml"
    click losses_config.yaml "https://github.com/barasamit/quantattack/blob/main/configs/losses_config.yaml"
    click model_config.yaml "https://github.com/barasamit/quantattack/blob/main/configs/model_config.yaml"

    click data_utils.py "https://github.com/barasamit/quantattack/blob/main/utils/data_utils.py"
    click map_imagnet.py "https://github.com/barasamit/quantattack/blob/main/utils/map_imagnet.py"
    click image_net_map.json "https://github.com/barasamit/quantattack/blob/main/utils/image_net_map.json"

    click load_RepQ_ViT.py "https://github.com/barasamit/quantattack/blob/main/load_RepQ_ViT.py"
    click load_ptq4vit.py "https://github.com/barasamit/quantattack/blob/main/load_ptq4vit.py"
    click model_utils.py "https://github.com/barasamit/quantattack/blob/main/utils/model_utils.py"

    click attack.py "https://github.com/barasamit/quantattack/blob/main/attack.py"
    click projected_gradient_descent.py "https://github.com/barasamit/quantattack/blob/main/attacks/projected_gradient_descent.py"
    click attack_utils.py "https://github.com/barasamit/quantattack/blob/main/utils/attack_utils.py"

    click losses.py "https://github.com/barasamit/quantattack/blob/main/losses.py"
    click losses_utils.py "https://github.com/barasamit/quantattack/blob/main/utils/losses_utils.py"
    click defenses.py "https://github.com/barasamit/quantattack/blob/main/utils/defenses.py"

    click clean_vs_quant.py "https://github.com/barasamit/quantattack/blob/main/measures/clean_vs_quant.py"
    click single_eval_only.py "https://github.com/barasamit/quantattack/blob/main/single_eval_only.py"
    click combine_csvs.ipynb "https://github.com/barasamit/quantattack/blob/main/notebooks/combine_csvs.ipynb"

    %% Styles
    classDef cli fill:#AED6F1,stroke:#2E86C1;
    classDef config fill:#D6EAF8,stroke:#1B4F72;
    classDef data fill:#D5F5E3,stroke:#196F3D;
    classDef compute fill:#F5B7B1,stroke:#A93226;
    classDef eval fill:#F9E79F,stroke:#B7950B;

    class ManyToMany,OneToOne,Universal,CLI_SingleEval cli
    class AttackConfig,AttacksConfig,DatasetClassConfig,DatasetConfig,LossesConfig,ModelConfig config
    class DataUtils,MapImageNet,ImageNetMap data
    class LoadRepQ,LoadPTQ,ModelUtils data
    class BaseAttack,PGDAttack,AttackUtils compute
    class Losses,LossUtils,Defenses compute
    class CleanVsQuant,EvalSingle,CombineCSV eval
```
