This package demonstrates a speed-first public verification profile for a user-provided Qwen3.6 35B-A3B compatible GGUF model on consumer 8GB-class VRAM hardware.

Public profile:

Qwen3.6-35B-A3B-G-Rotary

Reference local observation:

35.11 tok/s

The measurement is a local observation, not a universal guarantee.

Download

Use:

G-Rotary-Qwen36-Single-Public-Verification-20260526.zip

Requirements

Windows 11
NVIDIA RTX-class 8GB VRAM GPU or compatible environment
Recommended 32GB system memory
compatible user-provided Qwen3.6 35B-A3B GGUF model
local llama-cli.exe

Quick Start

Unzip the package:

powershell

.\bin\g-rotary-run.exe --setup --model "D:\models\Qwen3.6-35B-A3B.gguf" --llama-cli "D:\tools\llama-cli.exe"
.\bin\g-rotary-run.exe --model "D:\models\Qwen3.6-35B-A3B.gguf" --mode single --thinking off --prompt "Hello"
.\bin\g-rotary-run.exe --model "D:\models\Qwen3.6-35B-A3B.gguf" --mode single --thinking on --prompt "Think through this problem."

Not Included

model weights
training data
private operating-point logic
private validation traces
internal build files

License

G-Rotary execution package: see LICENSE-G-ROTARY.
Qwen base model, not included: see LICENSE-QWEN.
License scope separation: see NOTICE.md.

Patent Notice

This verification package relates to patent-filed technology by Myeong Jun Jo. It is provided as an observable public verification artifact, not as an internal implementation disclosure.

myeongjun77/Qwen3.6-35B-A3B-G-Rotary

Qwen3.6-35B-A3B G-Rotary

What This Shows