zen-family/docs/content/index.mdx

---
title: Zen AI Documentation
description: Ultra-efficient language models from 600M to 480B parameters
---

import { Cards, Card } from 'fumadocs-ui/components/card';
import { Tab, Tabs } from 'fumadocs-ui/components/tabs';
import { Callout } from 'fumadocs-ui/components/callout';
import { Steps, Step } from 'fumadocs-ui/components/steps';

# Welcome to Zen AI

<Callout type="info">
  **v1.0.1 Released!** Security updates, documentation improvements, and enhanced zoo-gym integration. [Learn more →](/updates/v1.0.1)
</Callout>

Zen AI delivers state-of-the-art language models ranging from **600M to 480B parameters**, optimized for edge deployment with **95% less energy** consumption than traditional models.

## 🚀 Quick Start

<Tabs>
  <Tab label="Python">
    ```python
    from transformers import AutoModelForCausalLM, AutoTokenizer
    
    # Load any Zen model
    model = AutoModelForCausalLM.from_pretrained("zenlm/zen-eco")
    tokenizer = AutoTokenizer.from_pretrained("zenlm/zen-eco")
    
    # Generate text
    inputs = tokenizer("Hello, ", return_tensors="pt")
    outputs = model.generate(**inputs, max_length=50)
    print(tokenizer.decode(outputs[0]))
    ```
  </Tab>
  <Tab label="Zoo-Gym">
    ```python
    from zoo_gym import ZooGym
    
    # Fine-tune with zoo-gym
    gym = ZooGym("zenlm/zen-eco")
    gym.train(
        dataset="data.jsonl",
        epochs=3,
        use_lora=True
    )
    ```
  </Tab>
  <Tab label="CLI">
    ```bash
    # Install Zen CLI
    pip install zen-ai
    
    # Run inference
    zen chat --model eco
    
    # Fine-tune
    zen train --model nano --data training.jsonl
    ```
  </Tab>
</Tabs>

## 📊 Model Family

<Cards>
  <Card 
    title="Zen-Nano" 
    description="600M params • Edge devices • 100 tok/s"
    href="/models/zen-nano"
  />
  <Card 
    title="Zen-Eco" 
    description="4B params • Balanced • 50 tok/s"
    href="/models/zen-eco"
  />
  <Card 
    title="Zen-Coder" 
    description="480B/35B MoE • Code gen • 91% HumanEval"
    href="/models/zen-coder"
  />
  <Card 
    title="Zen-Omni" 
    description="30B/3B MoE • Multimodal • Vision+Audio"
    href="/models/zen-omni"
  />
  <Card 
    title="Zen-Next" 
    description="80B/3B MoE • Ultra-sparse • 96% efficiency"
    href="/models/zen-next"
  />
</Cards>

## 🎯 Key Features

### Architecture Innovation
- **Qwen3 Base**: Latest architecture (September 2025)
- **MoE Models**: 90-96% parameter efficiency
- **GQA Optimization**: 75% memory reduction
- **Flash Attention 2**: 2-4x faster inference

### Training Excellence
- **Zoo-Gym Framework**: Official training toolkit
- **Recursive Improvement**: 94% effectiveness
- **LoRA Fine-tuning**: 0.3-2% trainable params
- **Multi-format Support**: GGUF, MLX, ONNX

### Deployment Flexibility
- **Edge Devices**: From smartphones to RPi
- **Local Execution**: Complete privacy
- **Cloud Optional**: Hybrid deployment
- **95% Energy Saving**: Sustainable AI

## 📈 Benchmarks

| Model | Params | MMLU | GSM8K | HumanEval | Speed |
|-------|--------|------|-------|-----------|-------|
| Nano | 600M | 42.3% | 28.1% | 18.2% | 100 t/s |
| Eco | 4B | 51.7% | 32.4% | 22.6% | 50 t/s |
| Coder | 35B* | 78.9% | 71.2% | 91.3% | 30 t/s |
| Omni | 3B* | 65.4% | 58.3% | 45.2% | 45 t/s |
| Next | 3B* | 87.3% | 92.1% | 84.6% | 70 t/s |

*Active parameters (MoE models)

## 🛠️ Installation

<Steps>
  <Step>
    ### Install Dependencies
    ```bash
    pip install transformers torch zoo-gym
    ```
  </Step>
  <Step>
    ### Download Model
    ```python
    from transformers import AutoModelForCausalLM
    model = AutoModelForCausalLM.from_pretrained("zenlm/zen-eco")
    ```
  </Step>
  <Step>
    ### Run Inference
    ```python
    outputs = model.generate(inputs, max_length=100)
    ```
  </Step>
</Steps>

## 🤝 Partnership

Built through collaboration between:

- **[Hanzo AI](https://hanzo.ai)** - Techstars '24, AI innovation
- **[Zoo Labs Foundation](https://zoo.ai)** - 501(c)(3), open AI infrastructure

## 📚 Documentation

<Cards>
  <Card 
    title="Getting Started" 
    description="Installation and quick start guide"
    href="/quickstart"
  />
  <Card 
    title="Training Guide" 
    description="Fine-tune with zoo-gym"
    href="/training/zoo-gym"
  />
  <Card 
    title="API Reference" 
    description="Complete API documentation"
    href="/api/python"
  />
  <Card 
    title="Whitepaper" 
    description="Technical deep dive"
    href="/papers/whitepaper"
  />
</Cards>

## 🌟 Community

- **GitHub**: [github.com/zenlm](https://github.com/zenlm)
- **HuggingFace**: [huggingface.co/zenlm](https://huggingface.co/zenlm)
- **Discord**: [discord.gg/zen-ai](https://discord.gg/zen-ai)
- **Email**: [email protected]

---

<Callout>
  **Need Help?** Check our [FAQ](/resources/faq) or join our [Discord](https://discord.gg/zen-ai) community.
</Callout>