IndraQuantum V5

Advanced Quantized-Linguistic Model Architecture

Status: V5 (Final) / Stable

IndraQuantum V5 is a Complex-Valued Neural Network (CVNN) designed to capture "Quantum Linguistic States" (Position = Phase, Meaning = Magnitude).

Key Features

• Holographic Loss: Minimizes Circular Variance to prevent phase collapse.
• Quantum FFN: Uses Symplectic Coupling (Forced Rotation) to bind real and imaginary components.
• Cloud Native: Ready for Azure/AWS/GCP training.

Directory Structure

IndraQuantum/
├── data/                     # Data loading
├── indra/                    # Core Library (Modules, Models, Losses)
├── training/                 # Training scripts & configs
├── utils/                    # Loggers & Cloud IO
├── scripts/                  # Tools (HF Conversion, Testing)
├── runs/                     # Experiment artifacts
└── docs/                     # Documentation

Quick Start

1. Installation

pip install -r requirements.txt

2. Verify Install

python scripts/test_model.py

3. Train (Local)

python training/train.py --config training/config_v5.yaml

4. Convert to HuggingFace

python scripts/convert_to_hf.py --checkpoint runs/<id>/final_model.pt --config training/config_v5.yaml --output_dir hf_model

Documentation

• Math & Theory
• V5 Design Spec

License

MIT