README.md

---
license: apache-2.0
language:
- fr
- en
base_model:
- Qwen/Qwen3.6-27B
pipeline_tag: text-generation
tags:
- leon
- parallactic-ai
- code-generation
- full-stack
- landing-page
- tailwind
- shadcn
- react
- nextjs
- app-generation
- vllm
- qwen3
library_name: transformers
---

# Leon v2.1

**27B · Full-Stack App Generation · Open Beta · by Parallactic AI**

Leon is a specialized language model fine-tuned from Qwen3.6-27B for **full-stack application and landing page generation**. Leon generates beautiful, production-ready React/Next.js code using modern UI libraries — out of the box.

> ⚠️ **Open beta.** Leon v2.1 is in public beta. Your outputs help train future versions. Errors are possible — always review generated code.

---

## What Leon Does

Leon specializes in generating **high-visual-fidelity, production-ready** frontend code:

- Multi-section landing pages
- Full-stack Next.js applications
- React components with animations
- Tailwind + shadcn/ui layouts
- Aceternity UI and Magic UI patterns
- Framer Motion animations
- Responsive, accessible code by default

Leon is **not a general assistant**. It is purpose-built for app generation.

---

## Component Stack Leon Knows

Leon is trained to use and combine these libraries intelligently:

| Library | Role | When Leon uses it |
|---------|------|-------------------|
| **Tailwind CSS** | Core styling | Always |
| **shadcn/ui** | Functional components | Buttons, forms, cards, dialogs |
| **Aceternity UI** | Bold visual sections | Heroes, backgrounds, 3D cards |
| **Magic UI** | Polished micro-interactions | Animated beams, text effects, borders |
| **Framer Motion** | Animations | Scroll reveals, hover effects, stagger |
| **Lucide React** | Icons | Throughout |

---

## Model Details

| Property | Value |
|----------|-------|
| Base model | Qwen/Qwen3.6-27B |
| Parameters | 27B |
| Format | SafeTensors |
| Specialization | Full-stack app & landing page generation |
| Primary output | React / Next.js / Tailwind code |
| Languages | French · English |
| License | Apache 2.0 |
| Context length | 8,192 tokens |

---

## Quick Start

### vLLM (recommended)

```bash
pip install vllm
```

```bash
python -m vllm.entrypoints.openai.api_server \
  --model Jesiel-AI/Leon-v2.1 \
  --host 0.0.0.0 \
  --port 8000 \
  --max-model-len 8192 \
  --max-num-seqs 256 \
  --gpu-memory-utilization 0.92 \
  --enable-prefix-caching \
  --disable-log-requests
```

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="token")

response = client.chat.completions.create(
    model="Jesiel-AI/Leon-v2.1",
    messages=[
        {
            "role": "system",
            "content": "You are Leon, a full-stack app generation model by Parallactic AI. Generate clean, production-ready React/Next.js code using Tailwind CSS, shadcn/ui, Aceternity UI, Magic UI, and Framer Motion."
        },
        {
            "role": "user",
            "content": "Generate a hero section for a SaaS landing page with an animated background and a CTA button."
        }
    ],
    max_tokens=2048,
    temperature=0.7,
    stream=True,
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="", flush=True)
```

### Transformers

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Jesiel-AI/Leon-v2.1"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {
        "role": "system",
        "content": "You are Leon, a full-stack app generation model by Parallactic AI. Generate clean, production-ready React/Next.js code using Tailwind CSS, shadcn/ui, Aceternity UI, Magic UI, and Framer Motion."
    },
    {
        "role": "user",
        "content": "Generate a pricing section with 3 tiers using shadcn/ui cards and Tailwind."
    }
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.7, do_sample=True)

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True))
```

---

## Architecture

Leon is designed to work as part of the following stack:

```
User prompt
     ↓
Leon v2.1 (code generation)
     ↓
Generated React/Next.js code
     ↓
GROTTE sandbox (EU-sovereign execution + preview)
```

---

## v0.5 Scope

Leon v2.1 (open beta) is focused on:

- ✅ Multi-section landing pages
- ✅ High visual fidelity (Aceternity + Magic UI patterns)
- ✅ shadcn/ui component library
- ✅ Framer Motion animations
- ✅ Responsive by default
- ✅ Copy-paste ready output

Deferred to later versions:
- 3D elements (Three.js / Spline)
- Tool calling / MCP
- Design system ingestion

---

## Limitations

- Open beta — expect rough edges and hallucinations
- Always review generated code before deploying to production
- 3D and advanced tool-calling not yet supported
- Best results with clear, specific prompts describing sections and style

---

## Citation

```bibtex
@misc{leon2026,
  author       = {Rombley, Jesiel and Parallactic AI},
  title        = {Leon v2.1: A Specialized Full-Stack App Generation Model},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Jesiel-AI/Leon-v2.1}},
}
```