---
license: agpl-3.0
library_name: sentence-transformers
tags:
  - openeden
  - codebook
  - vector-quantization
  - bilingual
---

# OpenEden Codebook Base Model

This repository stores the OpenEden 8D physiological vector to semantic
codebook alignment model. It is released under AGPL-3.0 together with the
OpenEden runtime.

## Contents

- `text_encoder/`: fine-tuned `Qwen/Qwen3-Embedding-0.6B` SentenceTransformer
- `vector_projector.pt`: 8D vector projector into the text embedding space
- `metadata.json`: training configuration
- `evaluation.json`: retrieval evaluation metrics
- `local-model-artifact.json`: JVM runtime artifact used by the local CLI
- `LICENSE`: GNU Affero General Public License v3.0

## Training

- samples: 2048
- codebook nodes: 512
- base model: `Qwen/Qwen3-Embedding-0.6B`
- epochs: 5
- batch size: 16
- text encoder: unfrozen

## Evaluation

- exactTop1: 0.12939453125
- nodeTop1: 0.509765625
- nodeTop5: 0.65478515625

The stored vector dimensions are exactly `[L, P, E, S, tau, V, M, F]`.
Derived dissonance is not stored in this model or training data.