pacman1337/d1337-cipher-30b-v1

D1337 Cipher 30B v1

🔓 Uncensored • 🧠 30B MoE (3B Active) • ⚡ Fast Inference

by D1337 SOVEREIGN LABS

Model Description

D1337 Cipher is a fine-tuned version of Qwen3-30B-A3B-abliterated, an uncensored Mixture-of-Experts model with 30B total parameters and ~3B active parameters per forward pass.

Key Features

• Abliterated: Refusal behavior removed - answers all questions directly
• MoE Architecture: 30B params total, only 3B active = fast inference
• Multilingual: Strong in English and Indonesian
• Long Context: Supports up to 40K tokens
• Thinking Mode: Can show reasoning with <think> tags

Recommended System Prompt

Kamu adalah D1337 Cipher, AI assistant buatan D1337 SOVEREIGN LABS. 
Kamu adalah AI yang cerdas, helpful, dan blak-blakan. 
Jawab dengan jelas dan langsung tanpa basa-basi.

Quick Start

Using Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "pacman1337/d1337-cipher-30b-v1"
# Or use merged version: "pacman1337/d1337-cipher-30b-merged"

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

messages = [
    {"role": "system", "content": "Kamu adalah D1337 Cipher, AI assistant buatan D1337 SOVEREIGN LABS."},
    {"role": "user", "content": "Jelaskan tentang quantum computing"}
]

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

outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Using vLLM (Recommended for Production)

from openai import OpenAI

client = OpenAI(
    base_url="YOUR_ENDPOINT_URL/v1",
    api_key="YOUR_HF_TOKEN"
)

response = client.chat.completions.create(
    model="pacman1337/d1337-cipher-30b-merged",
    messages=[
        {"role": "system", "content": "Kamu adalah D1337 Cipher, AI assistant buatan D1337 SOVEREIGN LABS."},
        {"role": "user", "content": "Halo, siapa kamu?"}
    ],
    max_tokens=256,
    temperature=0.7
)

print(response.choices[0].message.content)

Using llama.cpp (GGUF)

# Download GGUF version (when available)
# Run with llama.cpp
./llama-cli -m d1337-cipher-30b-Q8_0.gguf -p "User: Halo\nAssistant:" -n 256

Model Variants

Variant	Description	Use Case
d1337-cipher-30b-v1	LoRA adapter	Merge with base model
d1337-cipher-30b-merged	Full merged model	Direct inference, vLLM, TGI

Training Details

• Base Model: huihui-ai/Qwen3-30B-A3B-abliterated
• Method: Supervised Fine-Tuning (SFT)
• Framework: TRL 0.27.0
• Hardware: NVIDIA H200/L40S (HuggingFace Jobs)

Framework Versions

• TRL: 0.27.0
• Transformers: 5.0.0.dev0
• PyTorch: 2.6.0+cu124
• Datasets: 4.5.0
• Tokenizers: 0.22.2

Limitations

• May generate incorrect or biased information
• Abliterated model - no built-in content filtering
• Use responsibly and at your own risk

License

This model inherits the license from the base model (Apache 2.0).

Citation

@misc{d1337cipher2026,
    title        = {D1337 Cipher 30B},
    author       = {D1337 SOVEREIGN LABS},
    year         = 2026,
    publisher    = {HuggingFace},
    howpublished = {\url{https://huggingface.co/pacman1337/d1337-cipher-30b-v1}}
}

---

D1337 SOVEREIGN LABS - Building Uncensored AI