# ml-security-research-lab/deepcompress-vit architecture

> Architecture diagram of the GitHub repository ml-security-research-lab/deepcompress-vit, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2025-09-25). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/ml-security-research-lab/deepcompress-vit
- Repository: https://github.com/ml-security-research-lab/deepcompress-vit

## Overview

##### Project type and purpose
- DeepCompress-ViT is a research codebase for compressing and fine-tuning Vision Transformer (ViT) models and running inference at the edge.
- It provides two CLI entrypoints: one for training compressed models (main.py) and one for inference/evaluation (inference.py).
- It uses PyTorch/timm, implements quantization, autoencoder-based compression, and distillation.

##### File structure
- Top level:
- main.py – training CLI driver
- inference.py – inference/evaluation CLI driver
- requirements.txt – dependency list
- saved_models/ (populated at runtime with checkpoints)
- src/
- models/ – definitions of ViT, autoencoder, quantization, encoder modules
- training/engine.py – training loop, optimizer, scheduler, distillation logic
- utils/ – data loading, metrics calculation, model-save/load helpers

##### README for additional insights
- Training steps: select model_name, rank, distillation weight, mixed_precision, device, etc.
- Checkpoint storage on Google Drive and local file system.
- Inference supports different datasets (ImageNet vs. CIFAR-10).

## Diagram (Mermaid)

```mermaid
flowchart TB
    %% Subgraphs for Training and Inference
    subgraph "Training Flow"
        direction TB
        CLI_TRAIN["main.py"]:::cli
        Data_Loader["Data Ingestion & Preprocessing\n(src/utils/data_utils.py)"]:::data
        subgraph "Model Components"
            direction TB
            Vit["Vision Transformer\n(src/models/vit.py)"]:::compute
            AutoEnc["Autoencoder Compressor\n(src/models/autoencoder.py)"]:::compute
            QuanBase["_quan_base.py"]:::compute
            Quant["Quantization Logic\n(src/models/quantization.py)"]:::compute
            Encoders["Encoder Modules\n(src/models/encoders.py)"]:::compute
        end
        Engine["Training Engine\n(src/training/engine.py)"]:::compute
        MetricsLog["Metrics & Logging\n(src/utils/metrics.py, src/utils/model_utils.py)"]:::data
        CheckStore["Checkpoint Store\n(saved_models/)"]:::storage
        GDrive["Google Drive\n(optional external)"]:::external

        CLI_TRAIN -->|"args, config"| Data_Loader
        Data_Loader -->|"batches"| Vit
        Data_Loader -->|"batches"| AutoEnc
        Vit -->|"model init"| Engine
        AutoEnc -->|"compressed model"| Engine
        QuanBase -->|"quant utils"| Quant
        Quant -->|"quantized model"| Engine
        Encoders -->|"encoder modules"| Engine
        Engine -->|"metrics"| MetricsLog
        Engine -->|"checkpoint"| CheckStore
        CheckStore -->|"sync"| GDrive
    end

    subgraph "Inference Flow"
        direction TB
        CLI_INF["inference.py"]:::cli
        LoadCP["Load Checkpoint\n(saved_models/)"]:::storage
        CLI_INF -->|"args, checkpoint path"| LoadCP
        LoadCP -->|"model state"| Vit
        LoadCP -->|"model state"| Quant
        CLI_INF -->|"dataset"| Data_Loader
        Data_Loader -->|"batches"| Vit
        Data_Loader -->|"batches"| Quant
        Vit -->|"predictions"| MetricsLog
        Quant -->|"predictions"| MetricsLog
        MetricsLog -->|"results"| CLI_INF
    end

    %% Click Events
    click CLI_TRAIN "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/main.py"
    click CLI_INF "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/inference.py"
    click Data_Loader "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/utils/data_utils.py"
    click Vit "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/models/vit.py"
    click AutoEnc "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/models/autoencoder.py"
    click QuanBase "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/models/_quan_base.py"
    click Quant "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/models/quantization.py"
    click Encoders "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/models/encoders.py"
    click Engine "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/training/engine.py"
    click MetricsLog "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/utils/metrics.py"
    click MetricsLog "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/src/utils/model_utils.py"
    click CLI_TRAIN "https://github.com/ml-security-research-lab/deepcompress-vit/blob/main/requirements.txt"

    %% Styles
    classDef cli fill:#cccccc,stroke:#333,stroke-width:1px;
    classDef data fill:#9fdf9f,stroke:#333,stroke-width:1px;
    classDef compute fill:#87cefa,stroke:#333,stroke-width:1px;
    classDef storage fill:#ffcc66,stroke:#333,stroke-width:1px;
    classDef external fill:#f4a261,stroke:#333,stroke-width:1px;
```
