wyolo 1.0.1

Professional YOLO training library with MLOps integration

wyolo: Professional YOLO MLOps Orchestrator

wyolo is an enterprise-grade framework designed to manage the full lifecycle of YOLO and RT-DETR models. By leveraging a robust State Machine architecture and native MLOps integrations, wyolo provides a high-resiliency environment for computer vision training, tracking, and deployment.

---

🏗️ Architecture & System Design

1. 🚶 Execution Walkthrough (State Machine)

The system operates as a deterministic pipeline governed by conditional logic. It ensures environment integrity before committing expensive GPU resources.

For a visual flowchart, please refer to README_GIT.md.

2. 🗺️ Detailed System Workflow

Sequence of operations between the orchestrator, specialized states, and external MLOps entities.

For the sequence interaction diagram, please refer to README_GIT.md.

3. 🗺️ Architecture Components

A layered view of the ecosystem, illustrating the separation between orchestration, business logic, and infrastructure.

For the architectural mindmap, please refer to README_GIT.md.

---

✨ Key Features & Performance

• 🚀 State-Driven Orchestration: Complex logic implemented via wpipe for maximum resiliency.
• 🛡️ Quality-First Design: Maintains Pylint > 9.5 and regular Bandit security audits.
• 📊 Native MLOps: Built-in MLflow tracking, automated model registry, and DVC data versioning.
• ⚡ Resource Efficiency: Real-time monitoring of Peak RAM, CPU usage, and VRAM optimization.
• 📦 Distributed Architecture: Designed to run as an independent worker container (wtrain-service).

---

⚙️ Lifecycle Management

a. Build Process (CI/CD)

1. Multi-Stage Dockerfile: Optimizes image size while providing CUDA/CUDNN support.
2. Environment Isolation: Automatic resolution of complex dependencies (RT-DETR, MLflow, Redis).
3. Sanity Checks: Static analysis executed during the build phase.

b. Runtime Process (Execution)

1. Initialization: train_service.sh triggers the service wrapper.
2. Discovery: Automatic discovery of GPU topology and dataset paths.
3. Orchestration: The Pipeline engine manages retries, timeouts, and state transitions.
4. Persistence: Results and logs are persisted to wtrain.db (SQLite WAL) for audit trails.

---

📂 File-by-File Guide

Component	Description
src/wyolo/app/main.py	Entry point. Configures the wpipe state machine and orchestrates the worker.
src/wyolo/app/states/	Atomic state implementations (Check GPU, Dataset, MinIO, Train).
src/wyolo/core/	Core logic for the TrainerWrapper and MLflowManager.
src/wyolo/docker/	Production-ready orchestration scripts and requirements.
src/wyolo/trainer/	Specialized DTOs and implementation of the Elemental design pattern.
Makefile	The central command center for installation, testing, and deployment.
index.html	High-impact landing page for project stakeholders.

---

📂 Project Structure

src/wyolo
├── app
│   ├── main.py                <-- Orchestrator
│   └── states                 <-- State Machine Steps
│       ├── check/             <-- Validation Logic
│       ├── train/             <-- Training Execution
│       └── error_process/     <-- Resiliency Handling
├── core
│   ├── trainer_wrapper.py     <-- Engine Wrapper
│   └── mlflow_manager.py      <-- MLOps Tracking
└── docker
    ├── train_service.sh       <-- Production Entrypoint
    └── requirements.txt       <-- Locked Dependencies

---

🚀 Installation & Usage

# 1. Setup environment
make install

# 2. Run linting & security
make lint

# 3. Execute training suite
wyolo-train --config_path my_config.yaml

---

👨‍💻 Author

William Rodríguez - wisrovi
Technology Evangelist & AI Solutions Architect
LinkedIn Profile

---

📄 Bibliography & Resources

• Ultralytics Framework
• MLflow MLOps Platform
• wpipe Orchestration Library