🧬 GAJE Helix — Official Sovereign Genomic Models
Official model repository for the GAJE (Genetic Adaptive Joint Embedding) semantic compression and genomic neural memory platform.
🌟 What is a .gaje Model?
A .gaje file (Format Version v2312) is an all-in-one sovereign binary container:
- 🚀 Zero-Copy Mmap: Directly projected into virtual memory by the OS kernel without loading passes.
- 📚 Embedded Native GTOK Tokenizer: Contains its own BPE vocabulary and merge rules, eliminating mismatch and external JSON tokenizers.
- ⚡ Quantized Genomic Weights: Native Q4_0 / Q2_0 tensor layout optimized for AVX-512, AVX2, NEON, and WASM SIMD128.
- 🏝️ Island Model Compatible: Natively supports
.gmemepisodic, documental, and conversational memory islands.
📦 Available Models
| Model | Arch | Hidden / Layers | Format | Size | Description |
|---|---|---|---|---|---|
max.gaje |
Llama | 256 / 8L | Q2_0 + GTOK | ~99.5 MB | Flagship Organism: Ultra-fast, minimal memory, autonomous edge agent |
🚀 Quick Start with gaje-cli
Install and run locally in single-binary Rust (Zero-Python runtime):
# 1. Download the flagship model
gaje-cli pull max
# 2. Interactive Terminal Chat
gaje-cli chat --model models/born/max.gaje
# 3. Launch the Sovereign Web UI Server (Streaming SSE)
gaje-cli serve --model models/born/max.gaje --port 8080 --chat-only
🌐 Live Web Environments
- Hugging Face Space: huggingface.co/spaces/eaguilar/gaje-helix-web
- Vercel Production PWA: gaje-web-ui.vercel.app
📄 License
This repository and its models are distributed under the GNU Affero General Public License v3.0 (AGPL-3.0).