Pungyo verification package

LICENSE-LLAMACPP

@@ -0,0 +1,23 @@
+MIT License
+
+Copyright (c) 2023-2024 The ggml authors (https://github.com/ggml-org/ggml/blob/master/AUTHORS)
+Copyright (c) 2023-2024 The llama.cpp authors (https://github.com/ggml-org/llama.cpp/blob/master/AUTHORS)
+Copyright (c) 2024-2025 The ik_llama.cpp authors (https://github.com/ikawrakow/ik_llama.cpp/blob/main/AUTHORS)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

NOTICE.md

@@ -20,7 +20,9 @@ holders.
 
 ## 3. Upstream Runtime
 
-- Built with reference to llama.cpp / ik_llama.cpp (MIT License).
+- This package's executable is built from llama.cpp / ik_llama.cpp (MIT License).
+  The full upstream MIT license and copyright notices are included as
+  `LICENSE-LLAMACPP` per the MIT terms.
 - The Pungyo package adds an execution layer and does not relicense the
   upstream runtime.
 

Pungyo-Qwen36-Single-Public-Verification-20260530.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:584555f932cd6f7b0e89ef32d2a8253e3b368348bdc38d354b5b2060ed459da3
+size 597490045

README.md

@@ -17,26 +17,25 @@ library_name: gguf
 
 # Pungyo (풍요) — Run Qwen3.6-35B-A3B on an 8GB laptop
 
-✅ **It works.** A 35B-A3B class model runs on consumer 8GB-class VRAM.
+> **램을 풍요롭게 — Run big. Keep your memory abundant.**
+
+✅ **It works.** Single self-contained binary. Provide a Qwen GGUF, run, done.
 
 This repository is **not a model release**.
 Model weights are **not included**.
 The internal mechanism is **not disclosed**.
 
-## What This Shows
-
-Pungyo is a public verification package that demonstrates running a
-user-provided Qwen3.6 35B-A3B compatible GGUF model on a single consumer
-8GB-class GPU.
-
-Reference local observation:
+## Reference local observation (single profile)
 
 ```text
-34.5 tok/s   (8GB laptop GPU, ~7.1GB VRAM, warm; peak 35.5 tok/s)
+33.27 tok/s   (8GB laptop GPU, ~7GB VRAM, steady-state)
 ```
 
-Measured locally on Qwen3.6-35B-A3B (Q4_K_M, user-provided GGUF), 2026-05-29.
-The measurement is a local observation, not a universal guarantee.
+Measured on Qwen3.6-35B-A3B (Q4_K_M, user-provided GGUF), 2026-05-30, 245-token generation.
+A local observation, not a universal guarantee.
+
+> Want maximum memory headroom / stability? See the twin profile:
+> **myeongjun77/Pungyo-Qwen36-A3B-Twin**
 
 ## How does it work?
 
@@ -44,44 +43,40 @@ Not explained here. The mechanism is covered by patent-filed technology
 (application filed, KR). This release only demonstrates **that it works** —
 it does not disclose **how**.
 
-If you want the principle, it is in the inventor's patent and forthcoming
-paper. This package is an observable artifact, nothing more.
+## Download & Run (3 steps)
 
-## Download
+1. Get a compatible **Qwen3.6 35B-A3B GGUF** from https://huggingface.co/Qwen
+2. Download `Pungyo-Qwen36-Single-Public-Verification-20260530.zip` from this repo, unzip.
+3. Run:
 
-```text
-Pungyo-Qwen36-Single-Public-Verification-20260529.zip
+```powershell
+.\bin\pungyo.exe "path\to\Qwen3.6-35B-A3B.gguf" "Hello, my name is"
 ```
 
-## Requirements
-
-- Windows 11
-- NVIDIA RTX-class 8GB VRAM GPU (or compatible)
-- 32GB system memory recommended
-- a user-provided Qwen3.6 35B-A3B compatible GGUF model
-- local `llama-cli.exe` / `llama-server.exe`
+That's it. No setup, no separate llama.cpp install, no flags. The binary picks
+the operating point in-memory and runs in-process.
 
-## Quick Start
+## Requirements
 
-```powershell
-.\bin\pungyo.exe --setup --model "D:\models\Qwen3.6-35B-A3B.gguf" --llama-cli "D:\tools\llama-cli.exe"
-.\bin\pungyo.exe --model "D:\models\Qwen3.6-35B-A3B.gguf" --prompt "Hello"
-```
+- 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
 
-See `run.md` for details.
+(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
 
-- model weights
+- Qwen model weights (get from https://huggingface.co/Qwen)
 - training data
 - internal mechanism / operating-point logic
 - private validation traces
-- internal build files
 
 ## License
 
 - Pungyo execution package: see `LICENSE-PUNGYO`.
-- Qwen base model, not included: see `LICENSE-QWEN`.
+- 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

run.md

@@ -1,40 +1,37 @@
-# Pungyo — Run Guide
+# Pungyo — Run Guide (single profile)
 
-Pungyo runs a large Qwen3.6-35B-A3B class GGUF model on a single 8GB-class GPU.
-You provide the model. Pungyo provides the runner.
+Run a Qwen3.6 35B-A3B class GGUF on a single 8GB GPU. You provide the model.
+Pungyo is a single self-contained binary.
 
-## 1. Get a model (not included)
+## 1. Get a model
 
-Download a compatible Qwen3.6 35B-A3B GGUF from the official Qwen source:
+Download a compatible Qwen3.6 35B-A3B GGUF from the official source:
 https://huggingface.co/Qwen
 
 Pungyo does not ship, modify, or relicense model weights.
 
-## 2. Unzip Pungyo
+## 2. Unzip
 
 ```
-Pungyo-Qwen36-Single-Public-Verification-20260529.zip
-└── bin\pungyo.exe
+Pungyo-Qwen36-Single-Public-Verification-20260530.zip
+├── bin\pungyo.exe
+├── bin\cublas64_13.dll
+└── bin\cublasLt64_13.dll
 ```
 
-## 3. Setup (one time)
+## 3. Run
 
 ```powershell
-.\bin\pungyo.exe --setup `
-    --model "D:\models\Qwen3.6-35B-A3B.gguf" `
-    --llama-cli "D:\tools\llama-cli.exe"
+.\bin\pungyo.exe "D:\models\Qwen3.6-35B-A3B.gguf" "Hello, my name is"
 ```
 
-## 4. Run
-
-```powershell
-.\bin\pungyo.exe --model "D:\models\Qwen3.6-35B-A3B.gguf" --prompt "Hello"
-```
+That is the full invocation. The residency operating point is decided by Pungyo
+in memory and never appears on the command line.
 
 ## Notes
 
-- Requires Windows 11, NVIDIA 8GB-class GPU, 32GB system RAM recommended.
+- Requires Windows, NVIDIA RTX-class 8GB GPU with a recent NVIDIA driver.
+- 32 GB system RAM recommended.
 - Speed is a local observation and varies by hardware.
-- The mechanism is not documented here by design (patent-filed). This guide
-  covers operation only.
+- The mechanism is not documented here by design (patent-filed).
 - Issues / requests: use the repository Community tab.