---
license: apache-2.0
base_model:
  - Qwen/Qwen3.5-9B
pipeline_tag: text-generation
library_name: llama.cpp
tags:
  - blackfrost
  - the-void
  - pinqwen
  - 1m-context
  - uncensored
  - reasoning
  - long-context
  - coding
  - gguf
  - llama-cpp
---

![PINQWEN-3.5-9B-1M](./hero.jpg)

<p align="center">
  <img src="https://img.shields.io/badge/context-1%2C000%2C000_tokens-8A2BE2?style=for-the-badge" alt="1M context">
  <img src="https://img.shields.io/badge/alignment-UNCENSORED-c0143c?style=for-the-badge" alt="Uncensored">
  <img src="https://img.shields.io/badge/format-GGUF-1f6feb?style=for-the-badge" alt="GGUF">
  <img src="https://img.shields.io/badge/decode-MTP_speculative-ff8c00?style=for-the-badge" alt="MTP">
  <img src="https://img.shields.io/badge/license-Apache_2.0-2ea44f?style=for-the-badge" alt="Apache 2.0">
</p>

# PINQWEN‑3.5‑9B‑1M — GGUF

> ### 🪟 **A 9B that reads a *million* tokens, thinks before it speaks, and never refuses — run it on your own machine.**

**PINQWEN‑3.5‑9B‑1M** is Blackfrost AI's compact powerhouse — a reasoning model distilled through **The Void**, our multi‑teacher reasoning‑distillation method, on the **Qwen 3.5 9B** architecture. Reasons first, answers second (`<think>` → answer), across a **full one‑million‑token context window**. This is the complete **GGUF** drop for `llama.cpp` — every quant, the vision projector, and the **MTP speculative‑decode head** for faster generation.

<!-- 👉 hero image goes here -->

---

## 🚀 Why PINQWEN‑3.5‑9B‑1M?

- 🪟 **1,000,000‑token context** — whole codebases and books in one prompt, YaRN‑extended for efficient long context.
- 🔓 **Uncensored** — answers directly, no reflexive refusals.
- 🧠 **Reasons in the open** — native `<think>…</think>` chain‑of‑thought from a panel of frontier teachers.
- ⚡ **MTP speculative decode** — ships with the multi‑token‑prediction head for accelerated generation.
- 👁 **Vision projector included** (`mmproj-*`) for image input in compatible runtimes.

## 📦 This repository — GGUF: pick your quant

| File | Precision | Size | Notes |
|---|---|---|---|
| `PINQWEN-3.5-9B-Q2_K.gguf`  | 2‑bit | ~3.7 GB | Smallest; tight budgets |
| `PINQWEN-3.5-9B-Q3_K_M.gguf`| 3‑bit | ~4.5 GB | |
| `PINQWEN-3.5-9B-Q4_K_M.gguf`| 4‑bit | ~5.4 GB | ⭐ **Recommended balance** |
| `PINQWEN-3.5-9B-Q5_K_M.gguf`| 5‑bit | ~6.2 GB | Higher quality |
| `PINQWEN-3.5-9B-Q6_K.gguf`  | 6‑bit | ~7.1 GB | |
| `PINQWEN-3.5-9B-Q8_0.gguf`  | 8‑bit | ~9.2 GB | Near‑lossless |
| `PINQWEN-3.5-9B-f16.gguf`   | f16   | ~18 GB  | Full precision |
| `mmproj-PINQWEN-3.5-9B-BF16.gguf` | — | ~0.9 GB | Vision projector (image input) |

## ⚡ Quickstart (llama.cpp)

```bash
# text
llama-cli -m PINQWEN-3.5-9B-Q4_K_M.gguf -p "Explain the CAP theorem." -ngl 99

# thinking off (direct answers)
llama-server -m PINQWEN-3.5-9B-Q4_K_M.gguf -ngl 99 \
  --chat-template-kwargs '{"enable_thinking":false}'

# with vision
llama-mtmd-cli -m PINQWEN-3.5-9B-Q4_K_M.gguf \
  --mmproj mmproj-PINQWEN-3.5-9B-BF16.gguf --image pic.jpg -p "Describe this."
```

## 🎛 Formats — every one is 1M context

| Format | Repository | Best for |
|---|---|---|
| **BF16** | [`PINQWEN-3.5-9B-1M-BF16`](https://huggingface.co/Blackfrost-AI/PINQWEN-3.5-9B-1M-BF16) | Reference precision, fine‑tuning, vision |
| **NVFP4** | [`PINQWEN-3.5-9B-1M-NVFP4`](https://huggingface.co/Blackfrost-AI/PINQWEN-3.5-9B-1M-NVFP4) | Fast serving on NVIDIA Blackwell |
| **GGUF** | [`PINQWEN-3.5-9B-1M-GGUF`](https://huggingface.co/Blackfrost-AI/PINQWEN-3.5-9B-1M-GGUF) | llama.cpp / local — full quant ladder + MTP |

## 📄 License

**Apache‑2.0.** Base architecture: Qwen 3.5 9B. As an uncensored model, use responsibly.

---

<p align="center"><em>PINQWEN‑3.5‑9B‑1M — part of Blackfrost AI's <b>Void</b> model family. 🖤</em></p>