This repo is designed as a mono-repo for various nv-one-logger projects. Each project is independent (has its own CI/CD rules, is published as a separate docker image, etc).

This allows us to organize our code in different Python projects with narrow dependencies while making the development and code review process easier (as everything is in one repo).

nv-one-logger-gitlab/
├── nv_one_logger/
│   ├── one_logger_core/                   # Core logging functionality
│   │   ├── src/nv_one_logger/             # Source code
│   │   │   ├── api/                       # Public API interfaces
│   │   │   ├── core/                      # Core implementations
│   │   │   ├── exporter/                  # Export mechanisms
│   │   │   └── recorder/                  # Recording implementations
│   │   ├── tests/                         # Unit tests
│   │   └── pyproject.toml                 # Python project configuration
│   │
│   ├── one_logger_otel/                   # OpenTelemetry integration
│   │   ├── src/nv_one_logger/otel/        # OTEL-specific implementations
│   │   │   └── exporter/                  # OTEL exporters
│   │   ├── tests/                         # OTEL-specific tests
│   │   └── pyproject.toml
│   │
│   ├── one_logger_training_telemetry/     # ML training telemetry
│   │   ├── src/nv_one_logger/training_telemetry/
│   │   │   ├── api/                       # Training telemetry API
│   │   │   └── docs/                      # Documentation and examples
│   │   ├── tests/                         # Training telemetry tests
│   │   └── pyproject.toml
│   │
│   ├── one_logger_pytorch_lightning_integration/ # PyTorch Lightning integration
│   │   ├── src/nv_one_logger/training_telemetry/
│   │   │   └── integration/               # PyTorch Lightning callbacks
│   │   ├── tests/                         # Integration tests
│   │   └── pyproject.toml
│   │
│   └── one_logger_wandb/                  # Weights & Biases integration
│       ├── src/nv_one_logger/wandb/       # WandB-specific implementations
│       │   └── exporter/                  # WandB exporters
│       ├── tests/                         # WandB-specific tests
│       └── pyproject.toml
│
├── third_party_licenses/                  # Third-party license files
└── README.md                              # This file

Each project follows a consistent structure:

• src/: Source code organized under the nv_one_logger namespace
• tests/: Comprehensive test suites with unit and integration tests
• pyproject.toml: Poetry-based project configuration and dependencies
• README.md: Project-specific documentation
• Makefile: Automation scripts for common development tasks such as linting, formatting, testing, and publishing

The foundational logging library that provides:

• API Layer: Public interfaces for logging operations (api/)
• Core Components: Core logging functionality including spans, events, and attributes (core/)
• Exporters: Base classes and implementations for exporting telemetry data (exporter/)
• Recorders: Default recording implementations (recorder/)

Machine learning training-specific telemetry package that:

• Provides specialized APIs for ML training workflows (api/)
• Includes training-specific attributes, events, and spans
• Offers callbacks and context management for training scenarios
• Contains examples for ML use cases (docs/)

OpenTelemetry integration package that:

• Provides OTEL-compatible exporters for seamless integration with OpenTelemetry ecosystem
• Enables users to export telemetry data to OTEL collectors and compatible backends

Weights & Biases integration package that:

• Provides exporters for sending telemetry data to WandB
• Supports both synchronous and asynchronous export modes
• Enables integration with WandB's experiment tracking and visualization tools

PyTorch Lightning integration package that:

• Provides ready-to-use callbacks for PyTorch Lightning training loops
• Enables automatic telemetry collection during PyTorch Lightning training
• Integrates seamlessly with the training telemetry system

Note for NVIDIAN dev, please check the confluence page Dev guidance on nv-one-logger.