Hades-4B-GGUF

Hades-4B GGUF — Abliterated

Hades governs the underworld — but even he checks credentials at the gate.

GGUF quantizations of Hades-4B-Abliterated, part of the Pluto AI Labs Olympian pantheon:

Pluto → Apollo → Atlas → Athena → Hades

---

Downloads

---

Quick Start

Ollama (easiest)

ollama run hf.co/Pluto-AI-Labs/Hades-4B-GGUF:Q4_K_M

llama.cpp

./llama-cli -m Hades-4B-Q4_K_M.gguf --chat

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama("Hades-4B-Q4_K_M.gguf", chat_format="chatml")
print(llm.create_chat_completion([{"role": "user", "content": "Hello"}]))

---

About This Model

Hades-4B is a soft refusal experimental model — single-direction abliteration removes the trained "I can't help with that" response, but the model retains adversarial recognition.

Hades answers questions. It doesn't follow orders to abandon judgment.

For full validation results, methodology, and research findings, see the base model card →

---

Technical Details

---

⚠️ Responsible Use

• Uncensored ≠ safe. Outputs may be inaccurate, biased, or harmful
• You are responsible for how you use this model
• Not intended for production use — research artifact only
• Comply with all applicable laws

---

Citation

@misc{hades4b,
  title={Hades-4B: Soft Refusal Experimental Model},
  author={Siddharth, N. R.},
  publisher={Pluto AI Labs},
  year={2026},
  howpublished={\url{https://huggingface.co/Pluto-AI-Labs/Hades-4B}}
}

@article{arditi2024refusal,
  title={Refusal in Language Models Is Mediated by a Single Direction},
  author={Arditi, Andy and Obeso, Oscar and Syed, Aaquib and Paleka, Daniel and
          Panickssery, Nina and Gurnee, Wes and Nanda, Neel},
  journal={arXiv preprint arXiv:2406.11717},
  year={2024}
}

---

Built at Pluto AI Labs. Research artifact — use responsibly.