---
license: other
language:
- en
- ko
tags:
- local-inference
- gguf
- moe
- qwen
- windows
- consumer-gpu
- verification
pipeline_tag: text-generation
library_name: gguf
---

# Pungyo Twin (풍요 쌍발) — Run Qwen3.6-35B-A3B stably on an 8GB laptop

> **램을 풍요롭게 — Run big. Keep your memory abundant.**

✅ **It works, and it stays up.** Single self-contained binary. Stability-first
profile: more memory headroom, fewer out-of-memory failures under load.

This repository is **not a model release**.
Model weights are **not included**.
The internal mechanism is **not disclosed**.

## Single vs Twin (measured, 8GB laptop)

| profile | speed | VRAM | goal |
|---|---|---|---|
| Pungyo (single) | ~33.27 tok/s | ~7 GB (tight) | max speed |
| **Pungyo Twin** | **~28.50 tok/s** | **~3.7 GB (headroom)** | **stability** |

Twin nearly **halves VRAM pressure** (7 → 3.7 GB) for ~14% less speed, so the
model is far less likely to crash (OOM) during longer or heavier runs. Still
**one model**.

Measured on Qwen3.6-35B-A3B (Q4_K_M, user-provided GGUF), 2026-05-30,
245-token steady-state.

## How does it work?

Not explained here. The mechanism is covered by patent-filed technology
(application filed, KR). This release only demonstrates **that it works and
stays stable** — it does not disclose **how**.

## Download & Run (3 steps)

1. Get a compatible **Qwen3.6 35B-A3B GGUF** from https://huggingface.co/Qwen
2. Download `Pungyo-Qwen36-Twin-Public-Verification-20260530.zip` from this repo, unzip.
3. Run:

```powershell
.\bin\pungyo.exe "path\to\Qwen3.6-35B-A3B.gguf" "Hello, my name is" wireframe
```

The `wireframe` argument selects the twin (stability) profile. No setup, no
separate llama.cpp install. Operating point chosen in memory, never on the
command line.

## Requirements

- Windows 10/11
- NVIDIA RTX-class 8GB VRAM GPU (or compatible), CUDA driver installed
- 32 GB system memory recommended
- a user-provided Qwen3.6 35B-A3B compatible GGUF

(Bundled in the ZIP: `bin\pungyo.exe`, `bin\cublas64_13.dll`, `bin\cublasLt64_13.dll`. CUDA runtime libraries are included; you do NOT need the CUDA toolkit installed.)

## Not Included

- Qwen model weights (get from https://huggingface.co/Qwen)
- training data
- internal mechanism / operating-point logic
- private validation traces

## License

- Pungyo execution package: see `LICENSE-PUNGYO`.
- Built from llama.cpp / ik_llama.cpp (MIT): see `LICENSE-LLAMACPP`.
- Qwen base model (not included): see `LICENSE-QWEN`.
- License scope separation: see `NOTICE.md`.

## Patent Notice

Pungyo Twin relates to patent-filed technology by Myeong Jun Jo (application filed, KR).
It is provided as an observable public verification artifact, **not** as an
internal implementation disclosure. Commercial or patent licensing inquiries:
koing754@gmail.com