Instructions to use GilGamez-com/Leviathan-v4.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GilGamez-com/Leviathan-v4.2 with Transformers:

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

pipe = pipeline("text-generation", model="GilGamez-com/Leviathan-v4.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("GilGamez-com/Leviathan-v4.2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use GilGamez-com/Leviathan-v4.2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GilGamez-com/Leviathan-v4.2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GilGamez-com/Leviathan-v4.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GilGamez-com/Leviathan-v4.2

SGLang

How to use GilGamez-com/Leviathan-v4.2 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 "GilGamez-com/Leviathan-v4.2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GilGamez-com/Leviathan-v4.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "GilGamez-com/Leviathan-v4.2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GilGamez-com/Leviathan-v4.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GilGamez-com/Leviathan-v4.2 with Docker Model Runner:
```
docker model run hf.co/GilGamez-com/Leviathan-v4.2
```

🐉 Leviathan v4.2 — Enterprise Vision-Language Agent & Orchestrator

Leviathan v4.2 is a 27-billion parameter enterprise vision-language model and autonomous orchestrator developed by GilGamez. Fine-tuned for multi-step reasoning, secure software engineering, native Southeast Asian language understanding, and structured tool invocation, Leviathan v4.2 serves as a reliable operational core for enterprise automation and technical workflows.

⚡ Core Capabilities

Autonomous Tool Calling & Function Execution:
Natively trained to emit strict, schema-validated tool calls across search, sandboxed code execution, identity systems, transactional databases, and storage services.
Dual-Register Indonesian Fluency:
Natively understands and generates both formal Indonesian (Bahasa Baku adhering to PUEBI for executive reporting and enterprise compliance) and colloquial technical Indonesian (Bahasa Gaul for developer workflows and conversational agents).
Defensive Cybersecurity & Secure Coding:
Trained on secure design patterns, static vulnerability remediation (OWASP Top 10, CWE mitigation), and defensive systems hardening across PHP, JavaScript, Python, Go, and SQL.
Cognitive Invariant Defense & Anti-Sycophancy:
Engineered to reject leading questions, unverified premises, and conversational flattery traps. Preserves mathematical precision (such as IEEE 754 floating-point accuracy) and architectural invariants under user pressure.
Context Stability & Discussion Loop Closure:
Maintains long-horizon task awareness across multi-turn sessions, tracking agreed requirements without premature task abandonment or context degradation.

🛠 Model Specifications

Parameter	Specification
Model Name	Leviathan-v4.2
Developer	GilGamez
Base Architecture	Dense Vision-Language Transformer (27B Parameters)
Context Window	4,096 tokens (native training), scalable up to 32K+ via RoPE
Primary Languages	English, Bahasa Indonesia (Formal & Colloquial)
Task Scope	Autonomous Orchestration, Function Calling, Code Analysis, Systems Security
Format	LoRA Adapter (498 MB `safetensors`), GGUF (Available in GGUF repo)
License	Apache-2.0

📊 Training Methodology & Data Focus

Leviathan v4.2 was trained using Rank-Stabilized LoRA (RS-LoRA) with uniform embedding noise augmentation (NEFTune) on high-throughput accelerator hardware.

The training curriculum comprises curated, high-quality synthetic and human-verified domains:

Mathematical & Algorithmic Reasoning: Step-by-step chain-of-thought problem solving and formal logic.
Enterprise Full-Stack Engineering: Clean code architectures, modular API patterns, and database concurrency.
Defensive Systems Security: Vulnerability analysis, input sanitization, and secure-by-default software engineering.
Multi-Turn Tool Trajectories: Grounded function-calling dialogues with schema adherence and error recovery.
Regional & Multilingual Dialogue: High-fidelity Southeast Asian conversational corpora and technical terminology.

(Detailed internal training datasets and infrastructure specifications are proprietary to GilGamez).

💻 Quick Start & Usage

1. Python Inference (Transformers + PEFT)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "Qwen/Qwen3.8-27B"
adapter_id = "GilGamez-com/Leviathan-v4.2"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

model = PeftModel.from_pretrained(base_model, adapter_id)
model.eval()

messages = [
    {"role": "system", "content": "You are Leviathan v4.2, an autonomous AI assistant created by GilGamez."},
    {"role": "user", "content": "Explain how to mitigate Server-Side Request Forgery (SSRF) in high-throughput API gateways."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        temperature=0.6,
        top_p=0.9,
    )

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

2. High-Throughput Production Serving (vLLM)

vllm serve Qwen/Qwen3.8-27B \
    --enable-lora \
    --lora-modules leviathan-v4.2=GilGamez-com/Leviathan-v4.2 \
    --max-model-len 4096 \
    --gpu-memory-utilization 0.90 \
    --dtype bfloat16 \
    --enable-auto-tool-choice

3. Edge / Local Inference (Ollama)

Use the official Modelfile published in the repository:

ollama create leviathan-v4.2 -f Modelfile
ollama run leviathan-v4.2

🛡️ Safety & Alignment Principles

Leviathan v4.2 is trained strictly for defensive, educational, and enterprise operational purposes:

Defensive Focus: When asked about system vulnerabilities, Leviathan provides the theoretical mechanism followed immediately by production-grade mitigation code.
Zero Malicious Assistance: Refuses requests to generate automated exploitation payloads, malicious tooling, or evasion scripts.
Privacy by Design: Engineered to operate on sovereign private infrastructure, protecting confidential customer records and proprietary business logic.

🏛 Organization & Contact

Organization: GilGamez
Website: https://www.gilgamez.com
GGUF Repository: GilGamez-com/Leviathan-v4.2-GGUF

Citation

@misc{gilgamez2026leviathan42,
  title={Leviathan v4.2: Enterprise Vision-Language Orchestrator and Defensive Reasoning Engine},
  author={GilGamez AI Team},
  year={2026},
  publisher={GilGamez},
  howpublished={\url{https://huggingface.co/GilGamez-com/Leviathan-v4.2}}
}

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for GilGamez-com/Leviathan-v4.2

Base model

Qwen/Qwen3.8-27B

Quantized

unsloth/Qwen3.8-27B-unsloth-bnb-4bit

Finetuned

(8)

this model

Finetunes

1 model