Instructions to use MightyDragon-Dev/dragon_interceptor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MightyDragon-Dev/dragon_interceptor with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="MightyDragon-Dev/dragon_interceptor")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MightyDragon-Dev/dragon_interceptor")
model = AutoModelForCausalLM.from_pretrained("MightyDragon-Dev/dragon_interceptor", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use MightyDragon-Dev/dragon_interceptor with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MightyDragon-Dev/dragon_interceptor"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MightyDragon-Dev/dragon_interceptor",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/MightyDragon-Dev/dragon_interceptor

SGLang

How to use MightyDragon-Dev/dragon_interceptor with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "MightyDragon-Dev/dragon_interceptor" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MightyDragon-Dev/dragon_interceptor",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "MightyDragon-Dev/dragon_interceptor" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MightyDragon-Dev/dragon_interceptor",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use MightyDragon-Dev/dragon_interceptor with Docker Model Runner:
```
docker model run hf.co/MightyDragon-Dev/dragon_interceptor
```

dragon_interceptor / README.md

MightyDragon-Dev

Update README.md

088afc9 verified 4 months ago

preview code

Raw

History Blame Contribute Delete

2.92 kB

	---
	library_name: transformers
	tags:
	- generated_from_trainer
	model-index:
	- name: dragon_interceptor
	results: []
	---
	# 🐉 Dragon Interceptor (267k)

	Dragon Interceptor is an ultra-compact, high-speed generative model designed for aviation-inspired structural synthesis. Optimized for 35W mobile hardware, it bridges the gap between neural text generation and 2D pixel blueprints.

	[![Model Status: Active](https://img.shields.io/badge/Status-Active-brightgreen.svg)]()
	[![Hardware: i5-10210U](https://img.shields.io/badge/Hardware-Intel_i5--10210U-blue.svg)]()
	[![Parameters: 267k](https://img.shields.io/badge/Params-267k-orange.svg)]()

	---

	## 🚀 Performance HUD
	Benchmarks recorded on Intel Core i5-10210U (Surface Pro setup):

	* Weight Loading: 2055.47 it/s
	* Inference Speed: ~286 iterations/sec
	* Full Image Gen (28x28): ~0.88s
	* Brute Force Throughput: 1.2 - 2.5 images/sec (Turbo Mode)

	## 🛠️ Architecture
	The model utilizes a GPT-2 Causal LM backbone, repurposed for spatial data:
	- Vocab Size: 256 (Mapped to 8-bit grayscale intensity)
	- Sequence Length: 784 (Fixed $28 \times 28$ positional embeddings)
	- Parameters: 267,000 (Fits entirely within L2/L3 CPU cache)

	## 🛰️ Key Features

	### 1. Mosaic Synthesis
	By utilizing a sliding-window context bridge, the model can bypass its 784-position limit to generate seamless, multi-tile blueprints.
	- Tile Resolution: 28x28
	- Global Resolution: 112x112 (4x4 Mosaic) or custom strips.

	### 2. Seed Brute-Forcing
	The "Titan DNA" protocol allows for mass-generation of 1,000+ seeds to map the latent space for specific aircraft parts (wings, fuselages, tail fins).

	### 3. Thermal-Aware Inference
	Optimized for the Surface Pro's 35W power envelope. Uses raw Torch inference with KV-Caching to maintain stable frame rates even under thermal pressure.

	## 🖼️ Sample Inference & Visualization
	Use the following code to generate a high-fidelity "Radar Scan" from a specific seed. This snippet is optimized for 4K displays and technical clarity.

	```python
	import torch
	from transformers import AutoModelForCausalLM
	import matplotlib.pyplot as plt
	import os
	# Load Dragon Interceptor
	model = AutoModelForCausalLM.from_pretrained("MightyDragon-Dev/dragon_interceptor")

	# Generate a 28x28 Blueprint (Random Seed)
	seed_id = os.urandom(1)[0] % 10**6 # Random seed for variability
	print(f"🚀 Generating Dragon Blueprint with Seed {seed_id}...")
	input_ids = torch.tensor([[seed_id]])
	output = model.generate(input_ids, max_length=784, min_length=784, do_sample=True, temperature=0.7)

	# Reshape and Render
	blueprint = output[0].view(28, 28).detach().numpy()

	plt.figure(figsize=(8, 8), dpi=120)
	plt.imshow(blueprint, cmap='magma', interpolation='lanczos')
	plt.title(f"Dragon Interceptor: Sector Scan (Seed {seed_id})", color='white')
	plt.style.use('dark_background')
	plt.axis('off')
	plt.show()