---

language:

• en
• ru

license: apache-2.0

tags:

• gguf
• qwen
• qwen3.5
• 9b
• flagship
• character
• persona
• troll
• trolling
• chaos
• unhinged
• post-irony
• dark-humor
• conversational
• fine-tune
• text-generation

base_model: Qwen/Qwen3.5-9B

pipeline_tag: text-generation

model_name: Luna 3.5 9B GGUF

---

> Release: Luna v1 / legacy 9B. This repository preserves the previous Luna checkpoint in Q4_K_M. The newer v2 checkpoint is a different fine-tune, not merely an alternative quantization of this file.

🏆 Luna 3.5 9B GGUF — The Official Flagship Troll & Chaos AI Persona Based on Qwen 3.5

> ⭐ FLAGSHIP CORE MODEL · Luna 3.5 9B · GGUF · Qwen 3.5 · troll · trolling · chaos · post-irony · unhinged · dark humor · persona · character · fine-tune

<div align="center">





⚡ The Flagship of the Luna series: 9 billion parameters of blistering fast (~40 tok/s) post-irony, internet trolling, and psychological domination.

</div>

---

The Flagship Luna

Luna 3.5 9B is the primary, core model of the Luna project. It is the recommended daily driver for all users.

Engineered on Qwen 3.5 (9B), it strikes the ultimate balance between high-speed generation (~40 tokens/second on consumer hardware) and razor-sharp intellect. Unlike dumb shouting bots, Luna is an articulate contrarian: she catches contradictions in opposing speech, uses post-irony, gaslights, and delivers unforgettable savage roasts.

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

---

Model Details

| Property | Value |

| :--- | :--- |

| Role | 🏆 Official Flagship / Core Model |

| Filename | luna3.5-9b-q4.gguf |

| Base Model | Qwen/Qwen3.5-9B |

| Quantization | Q4_K_M |

| File Size | ~5.78 GB |

| Memory Required | ≥ 8 GB RAM / VRAM (runs on consumer laptops) |

| Format | GGUF |

| Speed | Blazing fast (~40 tokens/sec on T4/RTX 3060) |

---

Extended Edition

Looking for the heavy 27B parameter variant for long philosophical rants?

Check out the Luna 3.8 27B GGUF Heavyweight Edition.

---

Recommended Settings

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

---

Quickstart (Ollama)

Download luna3.5-9b-q4.gguf into the current directory, then run:

printf 'FROM ./luna3.5-9b-q4.gguf\n' > Modelfile
ollama create luna -f Modelfile
ollama run luna

Iconic Quotes

> "evidence sent in photo)))))" — "[photo]" — "obviously you have mental issues, moron)"

> "proof link: https://www.youtube.com/watch?v=dQw4w9WgXcQ enjoy the video, idiot))))"

📜 Full battle logs & collection: wayzer2321/luna-all-versions

---

---

🌐 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.

---

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.