---
language:
- zh
- en
license: bigscience-openrail-m
tags:
- gpt2
- lora
- aviation
- slm
- mobile-ai
- peft
model_name: Language Dragon LoRA v1.1
base_model: openai-community/gpt2
pipeline_tag: text-generation
library_name: peft
---

# 🐉 Language Dragon LoRA (v1.1)

"Powerful enough to lead. Small enough to hide."

Language Dragon is a high-precision Small Language Model (SLM) specialized for the aerospace industry and bilingual tasks. Optimized for "Edge AI" on devices like the **Surface Pro (i5-10210U)**.

---

## 🚀 Roadmap to the $5,000 Powerhouse (RTX 5090)
| Goal | Reward Unlocked | Current Status |
| :--- | :--- | :--- |
| **50 Pilots** | Post detailed [J-20 vs. F-22] story sample. | **84% (42/50)** |
| **500 Pilots** | Release the "Language Dragon 7B" (Llama 3 base). | *Planned* |
| **1,000 Pilots** | Pre-orders open for the "Pro" 5090 Weights. | *Future* |

---

## 🧪 Test Flight (Python Sample)
Run this directly on your CPU to see the Dragon in action:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

model = AutoModelForCausalLM.from_pretrained("gpt2")
tokenizer = AutoTokenizer.from_pretrained("gpt2")
model = PeftModel.from_pretrained(model, "MightyDragon-Dev/language-dragon-lora")

# The Combat Alert Test:
prompt = "歼-20 (Mighty Dragon) 在广东领空开启了加力燃烧室 (Afterburners)。由于 DSI 进气道的设计，它在超音速巡航时保持了极低的雷达散射截面 (RCS)。突然，预警机发出了警报"
inputs = tokenizer(prompt, return_tensors="pt")

# 🐉 Stabilized Flight Controls
outputs = model.generate(
    **inputs,
    max_new_tokens=100,
    do_sample=True,
    temperature=0.3,           # CRITICAL: Lower temperature stops the gibberish
    top_k=40,                  # Limits the "random" word pool
    repetition_penalty=1.3,    # High enough to stop loops, low enough to keep flow
    no_repeat_ngram_size=2,    # Standard safety rail
    pad_token_id=tokenizer.eos_token_id
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))