# skalskip/vlms-zero-to-hero architecture

> Architecture diagram of the GitHub repository skalskip/vlms-zero-to-hero, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2025-04-01). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/skalskip/vlms-zero-to-hero
- Repository: https://github.com/skalskip/vlms-zero-to-hero

## Overview

##### Project Type & Purpose
- Recognize that this is not a production application but a curated educational resource.
- It’s structured as a series of topic-specific tutorials (each in its own directory) that progressively introduce the reader to research papers and deep learning methods from traditional NLP/vision fundamentals to modern vision-language models.
- The roadmap in the README outlines a clear progression: NLP fundamentals → CV fundamentals → early vision-language models → scale/efficiency improvements → modern vision-language models → extra resources.

##### Breaking Down the File Structure
- Each top-level directory (e.g., “01_natural_language_processing_fundamentals”, “02_computer_vision_fundamentals”, etc.) represents a module or chapter in the educational series.

##### Main Components
- Educational Modules: Create separate blocks for each of the tutorial sections:
- NLP Fundamentals (e.g., Word2Vec tutorial notebook)
- Computer Vision Fundamentals
- Early Vision-Language Models
- Scale and Efficiency
- Modern Vision-Language Models
- Extra (additional evaluation methods like BLEU)
- Supporting Elements:
- Roadmap and tutorial resources (as described in the README) which serve as the backbone guiding the learning path.

##### Establishing Relationships and Flow
- How “extra” resources might intersect (or provide support) across the topics.
- Connections between the tutorial notebooks and their external references (papers, videos, colab links) to highlight the integration of theory and practice.
- Dependency flow – for instance, showing that understanding basics in NLP/CV is prerequisite knowledge for later modules covering modern VLMs.

##### Architectural Patterns & Design Principles
- Although it isn’t a full-stack application, think of this repository in terms of a “modular educational framework.” Each module is a self-contained unit that builds upon the previous ones.
- Foundational topics separately from advanced ones.
- Labels that clearly articulate the content scope (e.g., “NLP Fundamentals - Word2Vec Tutorial”, “Modern Vision-Language Models - BLIP-2 Overview”).

- At the top level, outline a high-level “roadmap” or flow chart that contains distinct blocks for each educational module.
- Below each block, add a layer to illustrate key notebooks or resources. For example, the “01_natural_language_processing_fundamentals” block would point to the specific notebook file and mention underlying frameworks (like PyTorch) if applicable.

## Diagram (Mermaid)

```mermaid
flowchart TD
    subgraph "Documentation"
        DOC_README["Project Roadmap & Documentation (README.md)"]:::doc
        DOC_LICENSE["Open Source License (LICENSE)"]:::doc
    end

    DOC_README --> FUND_NLP["NLP Fundamentals: Word2Vec Tutorial"]:::fundamental
    FUND_NLP --> FUND_CV["Computer Vision Fundamentals"]:::fundamental
    FUND_CV --> ADV_EARLY["Early Vision-Language Models"]:::advanced
    ADV_EARLY --> ADV_SCALE["Scale and Efficiency"]:::advanced
    ADV_SCALE --> ADV_MODERN["Modern Vision-Language Models"]:::advanced
    ADV_MODERN --> EXTRA["Extra Educational Resources"]:::extra

    click DOC_README "https://github.com/skalskip/vlms-zero-to-hero/blob/master/README.md"
    click DOC_LICENSE "https://github.com/skalskip/vlms-zero-to-hero/tree/master/LICENSE"
    click FUND_NLP "https://github.com/skalskip/vlms-zero-to-hero/blob/master/01_natural_language_processing_fundamentals/01_01_word2vec_with_sub_sampling_and_negative_sampling_in_pytorch.ipynb"
    click FUND_CV "https://github.com/skalskip/vlms-zero-to-hero/tree/master/02_computer_vision_fundamentals"
    click ADV_EARLY "https://github.com/skalskip/vlms-zero-to-hero/tree/master/03_early_vision_language_models"
    click ADV_SCALE "https://github.com/skalskip/vlms-zero-to-hero/tree/master/04_scale_and_efficiency"
    click ADV_MODERN "https://github.com/skalskip/vlms-zero-to-hero/tree/master/05_modern_vision_language_models"
    click EXTRA "https://github.com/skalskip/vlms-zero-to-hero/tree/master/06_extra"

    classDef fundamental fill:#AED581,stroke:#333,stroke-width:2px;
    classDef advanced fill:#81D4FA,stroke:#333,stroke-width:2px;
    classDef extra fill:#CE93D8,stroke:#333,stroke-width:2px;
    classDef doc fill:#FFB74D,stroke:#333,stroke-width:2px;
```
