soundbyte 0.1.0

Learn, Train, Deploy: An Academic-friendly DL Toolkit for Accelerated Learning and Prototyping

SoundByte – Learn, Train, Deploy: An Academic-friendly DL Toolkit for Accelerated Learning and Prototyping

A powerful, modular, and lightweight Python toolkit for training deep neural networks with minimal code and maximum flexibility.

🚀 Quick Start • 📖 Documentation • 🎯 Examples • 🤝 Contributing

---

✨ Key Features

🧩 Modular Design	📝 Minimal Code	⚙️ JSON Configuration	📊 Lightweight Dashboard	🔧 Easy Integration
Plug-and-play components for maximum flexibility	Run complex experiments with just a few lines of code	Control everything through intuitive JSON configs	Track experiments with built-in visualization	Seamlessly integrate custom models

🏗️ Ultra-Modular Architecture

• Mix & Match Components: Pre-built modules for data loading, model architectures, optimizers, and training loops
• Custom Pipeline Builder: Create unique training pipelines by combining modular components
• Plugin System: Extend functionality with custom plugins without modifying core code

🎯 Experiment-Ready Design

• One-Line Experiments: Launch complex training with a single command
• Hyperparameter Sweeps: Built-in support for automated hyperparameter optimization
• Reproducible Results: Automatic seed management and deterministic training

📋 JSON-Driven Configuration

• No Code Changes: Modify experiments entirely through JSON configuration files
• Schema Validation: Built-in validation ensures configuration correctness
• Template Library: Pre-made configs for common architectures and tasks

📈 Built-in Experiment Tracking

• Real-time Monitoring: Live training metrics and visualizations
• Model Comparison: Side-by-side comparison of different experiments
• Resource Monitoring: GPU/CPU utilization and memory tracking

🔗 Seamless Model Integration

• Auto-Discovery: Automatically detect and register custom model classes
• Multi-Framework Support: Works with PyTorch, Hugging Face Transformers, and more
• Zero-Boilerplate: Add new models with minimal wrapper code

---

🚀 Quick Start

Installation

# Install from PyPI (recommended)
pip install deepnet-toolkit

# Install with additional dependencies
pip install deepnet-toolkit[vision,nlp,audio]

# Install from source
git clone https://github.com/yourusername/deepnet-toolkit.git
cd deepnet-toolkit
pip install -e .

30-Second Example

from deepnet import Trainer, load_config

# Load configuration
config = load_config("configs/image_classification.json")

# Initialize trainer
trainer = Trainer(config)

# Train model
trainer.fit()

# Evaluate
results = trainer.evaluate()
print(f"Test Accuracy: {results['accuracy']:.2%}")

Configuration Example

{
  "experiment": {
    "name": "resnet50_cifar10",
    "seed": 42
  },
  "model": {
    "type": "ResNet50",
    "num_classes": 10
  },
  "data": {
    "dataset": "CIFAR10",
    "batch_size": 32,
    "num_workers": 4
  },
  "training": {
    "optimizer": "AdamW",
    "learning_rate": 1e-3,
    "epochs": 100,
    "scheduler": "CosineAnnealingLR"
  },
  "logging": {
    "dashboard": true,
    "save_checkpoints": true
  }
}

---

🏛️ Architecture Overview

graph TB
    A[JSON Config] --> B[Config Parser]
    B --> C[Data Module]
    B --> D[Model Module]
    B --> E[Trainer Module]
    B --> F[Logger Module]
    
    C --> G[DataLoader]
    D --> H[Neural Network]
    E --> I[Training Loop]
    F --> J[Dashboard]
    
    G --> I
    H --> I
    I --> J
    I --> K[Checkpoints]
    J --> L[Metrics & Visualizations]
    
    style A fill:#e1f5fe
    style B fill:#f3e5f5
    style I fill:#e8f5e8
    style J fill:#fff3e0

---

🎯 Examples

Image Classification

# Train ResNet on CIFAR-10
deepnet train --config configs/vision/resnet_cifar10.json

# Custom dataset
deepnet train --config configs/vision/custom_dataset.json --data.path ./my_dataset

Natural Language Processing

# Fine-tune BERT for text classification
deepnet train --config configs/nlp/bert_classification.json

# Custom tokenizer
deepnet train --config configs/nlp/custom_tokenizer.json --model.tokenizer ./my_tokenizer

Computer Vision

# Object detection with YOLO
deepnet train --config configs/vision/yolo_detection.json

# Semantic segmentation
deepnet train --config configs/vision/unet_segmentation.json

---

📊 Dashboard Preview

Real-time experiment tracking with built-in dashboard

Metrics Visualization	Model Comparison	Resource Monitoring
📈 Loss curves, accuracy plots	🔄 Side-by-side experiment comparison	💻 GPU/CPU utilization graphs
📊 Custom metric tracking	📋 Hyperparameter analysis	💾 Memory usage monitoring

---

🔧 Custom Model Integration

Adding your custom model is as simple as:

from deepnet import BaseModel
import torch.nn as nn

class MyCustomModel(BaseModel):
    def __init__(self, num_classes: int = 10, **kwargs):
        super().__init__()
        self.backbone = nn.Sequential(
            nn.Linear(784, 256),
            nn.ReLU(),
            nn.Linear(256, num_classes)
        )
    
    def forward(self, x):
        return self.backbone(x.view(x.size(0), -1))

# Auto-registration - no additional code needed!
# Use in config: {"model": {"type": "MyCustomModel", "num_classes": 10}}

---

📚 Documentation & Tutorials

Resource	Description
📖 Documentation	Complete API reference and guides
🎓 Tutorials	Step-by-step examples and best practices
🔬 Examples	Ready-to-run example scripts
📝 Configuration Guide	JSON configuration reference
🧩 Plugin Development	Create custom components

---

🛠️ Advanced Features

Multi-GPU Training

# Data parallel training
deepnet train --config config.json --gpus 0,1,2,3

# Distributed training
torchrun --nproc_per_node=4 deepnet/cli.py train --config config.json

Hyperparameter Optimization

{
  "hyperopt": {
    "method": "bayesian",
    "trials": 50,
    "search_space": {
      "training.learning_rate": ["log_uniform", 1e-5, 1e-1],
      "model.dropout": ["uniform", 0.1, 0.5],
      "training.batch_size": ["choice", [16, 32, 64]]
    }
  }
}

Custom Callbacks

from deepnet.callbacks import Callback

class CustomCallback(Callback):
    def on_epoch_end(self, epoch, logs):
        if logs['val_accuracy'] > 0.95:
            self.trainer.stop_training = True
            print("🎉 Target accuracy reached!")

---

🔗 Ecosystem Integrations

---

🚀 Performance Benchmarks

Model	Dataset	Training Time	Memory Usage	Accuracy
ResNet50	CIFAR-10	45 min	3.2 GB	94.2%
BERT-base	IMDB	2.1 hours	8.1 GB	91.8%
YOLOv5	COCO	8.3 hours	11.4 GB	65.1 mAP

Benchmarks run on NVIDIA RTX 3080, PyTorch 2.0

---

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Setup

# Clone repository
git clone https://github.com/yourusername/deepnet-toolkit.git
cd deepnet-toolkit

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Run linting
pre-commit run --all-files

🧪 Testing

# Run all tests
pytest tests/

# Run with coverage
pytest --cov=deepnet tests/

# Run specific test categories
pytest tests/test_models.py -v

---

📦 Installation Options

🐳 Docker Installation

# Pull the Docker image
docker pull yourusername/deepnet-toolkit:latest

# Run with GPU support
docker run --gpus all -it -v $(pwd):/workspace yourusername/deepnet-toolkit:latest

# Build from source
git clone https://github.com/yourusername/deepnet-toolkit.git
cd deepnet-toolkit
docker build -t deepnet-toolkit .

📦 Conda Installation

# Create conda environment
conda create -n deepnet python=3.8
conda activate deepnet

# Install from conda-forge
conda install -c conda-forge deepnet-toolkit

# Or install with pip in conda environment
pip install deepnet-toolkit

🔧 Development Installation

# Clone and install for development
git clone https://github.com/yourusername/deepnet-toolkit.git
cd deepnet-toolkit

# Install in editable mode with dev dependencies
pip install -e ".[dev,test,docs]"

# Install pre-commit hooks
pre-commit install

---

🏆 Comparison with Other Frameworks

Feature	DeepNet Toolkit	PyTorch Lightning	FastAI	Keras
JSON Configuration	✅ Full Support	❌ Limited	❌ No	❌ No
Zero-Code Experiments	✅ Yes	❌ Code Required	❌ Code Required	❌ Code Required
Built-in Dashboard	✅ Lightweight	❌ External Tools	❌ External Tools	❌ External Tools
Auto Model Discovery	✅ Yes	❌ Manual Registration	❌ Manual	❌ Manual
Multi-Framework Support	✅ PyTorch + HF	✅ PyTorch Only	✅ PyTorch Only	✅ TensorFlow Only

---

📜 License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

---

🙏 Acknowledgments

• Inspired by NVIDIA NeMo, ESPNet, and SpeechBrain
• Built with ❤️ using PyTorch and Hugging Face
• Special thanks to the open-source community

---

📞 Support & Community

Questions? Open an issue or start a discussion

Found a bug? Please report it on our issue tracker

Want to contribute? Check out our contribution guidelines

---

⭐ Star us on GitHub — it motivates us a lot!

Made with ❤️ by the DeepNet Team

⬆ Back to Top