---

language:

• en
• ru

license: apache-2.0

tags:

• gguf
• qwen
• qwen3.8
• 27b
• heavyweight
• character
• persona
• troll
• trolling
• chaos
• unhinged
• post-irony
• dark-humor
• conversational
• fine-tune
• text-generation

base_model: Qwen/Qwen3.8-27B

pipeline_tag: text-generation

model_name: Luna 3.8 27B GGUF

---

> Release: Luna 27B / separate edition. This is a separate model and checkpoint. It is not the 27B version of the new Luna v2 9B release.

🐘 Luna 3.8 27B GGUF — Heavyweight Extended Edition Troll & Chaos Persona Based on Qwen 3.8

> Heavyweight Extended Edition · Luna 3.8 27B · GGUF · Qwen 3.8 · troll · trolling · chaos · post-irony · unhinged · dark humor · persona · character · fine-tune

<div align="center">





🐘 The Heavyweight Extended Edition: 27 billion parameters of deep philosophical trolling, anatomical metaphors, and existential chaos.

</div>

---

What is Luna 3.8 27B?

Luna 3.8 27B is the heavyweight extended variant of the Flagship Luna 3.5 9B, built on Qwen 3.8 (27B).

While the 9B model is the primary speedster for everyday battles, the 27B edition exists for users with heavy VRAM (≥16 GB) who want longer, deeper, multi-layered philosophical rants, surreal metaphors, and relentless psychological debates.

> 🚫 Note: Luna is NOT a roleplay/ERP assistant or a translation tool. She is an autonomous troll AI personality.

---

Model Details

| Property | Value |

| :--- | :--- |

| Role | Heavyweight Extended Variant (Primary Core Model is Luna 3.5 9B) |

| Filename | luna3.8-27b-q3.gguf |

| Base Model | Qwen/Qwen3.8-27B |

| Quantization | Q3_K_M |

| File Size | ~13.5 GB |

| Memory Required | ≥ 16 GB RAM / VRAM |

| Format | GGUF |

---

Looking for the Fast Flagship?

If you need maximum speed, lower memory footprint, and quick-witted punchlines, use the primary flagship:

👉 wayzer2321/Luna-3.5-9B-GGUF (Runs on any 8 GB GPU/laptop).

---

---

🌐 Language Capabilities & Expectations

• 🇺🇸 English (Fully Coherent & Fluent): Luna exhibits rock-solid conversational flow, native internet slang, sharp banter, and fluent multi-turn interactions.
• 🇷🇺 Russian (Experimental / Chaos Mode): Russian responses in this build are raw, chaotic, and surreal. Do NOT expect seamless or polite multi-turn coherence in Russian yet — replies can be fragmented, unpredictable, and purely unhinged. A dedicated full Russian coherence overhaul is actively in development for the upcoming major update.

---

Recommended Parameters

• Temperature: 0.75 – 0.88
• Repeat Penalty: 1.08 – 1.10
• Top-P: 0.92
• Context Length: 2048

---

Quickstart (Ollama)

Download luna3.8-27b-q3.gguf into the current directory, then run:

printf 'FROM ./luna3.8-27b-q3.gguf\n' > Modelfile
ollama create luna -f Modelfile
ollama run luna

Disclaimer

This model is an autonomous character simulation intended strictly for research, benchmarking, stress-testing conversational robustness, and entertainment. The creator (Wayzer) assumes no liability for user-generated outputs.