Instructions to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
# Run inference directly in the terminal:
llama cli -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
# Run inference directly in the terminal:
llama cli -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
# Run inference directly in the terminal:
./llama-cli -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Use Docker

docker model run hf.co/CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

LM Studio
Jan
Ollama
How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with Ollama:
```
ollama run hf.co/CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
```
Unsloth Desktop

How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with Docker Model Runner:
```
docker model run hf.co/CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S
```

Lemonade

How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Run and chat with the model

lemonade run user.Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF-UD-IQ2_S

List all available models

lemonade list

Hermes Agent

How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF:UD-IQ2_S" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen3.8-27B UD-IQ2_S — RouteCache

Quality-passing UD-IQ2_S GGUF with a RouteCache execution profile embedded directly in GGUF metadata. Quantized tensor payload bytes are preserved from the recorded source GGUF; the whole-file SHA differs because the header metadata is new.

Ollama

ollama run hf.co/CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF

The HF→Ollama bridge reads the root-level params JSON in this repo. It contains {"num_ctx":4096}. Modelfile is also included for manual ollama create, but direct HF execution does not depend on it.

If you download/clone the entire Hugging Face repository on Windows, you can instead double-click RUN_OLLAMA.bat. It uses the local GGUF, installs Ollama through WinGet when available, creates qwen3.8-27b-routecache from the included Modelfile, and runs it.

Optimized RouteCache runtime

Use the companion GitHub repository and run SETUP_AND_RUN.bat for the hardware-specific RouteCache tensor placement, custom-kernel attempt, warm-up, low-latency streaming, and automatic rolling context compaction. OLLAMA_SETUP_AND_RUN.bat is the near-zero-step Ollama path.

Reproducibility

The GitHub repository includes REPRODUCE_MODEL.bat, which can rebuild the low-bit topology from official Qwen/Qwen3.8-27B weights using the pinned llama.cpp source plus Unsloth's public UD-IQ2_S importance matrix/per-tensor recipe. Source reproduction is optional and requires roughly 120–140 GiB temporary disk.

Calibration

Private RouteCache calibration on RTX 4060 8 GB: 4096 context, Q8_0 KV, batch/ubatch 256/64, explicit CPU embedding + FFN up/gate blocks 33–46, warmed non-MTP floor around 11.39 tok/s. Quality PPL ratio was about 1.057 against the local Q4 teacher (acceptance threshold 1.08). Hardware-specific speed claim only.

License

Model/quantization license identifier configured for publication: apache-2.0. Verify upstream attribution and any additional terms before publication.

Downloads last month: 1

GGUF

Model size

27B params

Architecture

qwen35

Hardware compatibility

2-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for CyberGamer/Qwen3.8-27B-UD-IQ2_S-RouteCache-GGUF

Base model

Qwen/Qwen3.8-27B

Quantized

(993)

this model