Instructions to use alphabell/alphaBell-Qwen3.8-27B-Open-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 alphabell/alphaBell-Qwen3.8-27B-Open-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 alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
# Run inference directly in the terminal:
llama cli -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
# Run inference directly in the terminal:
llama cli -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

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 alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
# Run inference directly in the terminal:
./llama-cli -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

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 alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
# Run inference directly in the terminal:
./build/bin/llama-cli -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Use Docker

docker model run hf.co/alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Llama new
LM Studio
Jan

vLLM

How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "alphabell/alphaBell-Qwen3.8-27B-Open-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "alphabell/alphaBell-Qwen3.8-27B-Open-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Ollama
How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with Ollama:
```
ollama run hf.co/alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
```
Unsloth Desktop

How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

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": "alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with Docker Model Runner:
```
docker model run hf.co/alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1
```

Lemonade

How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Run and chat with the model

lemonade run user.alphaBell-Qwen3.8-27B-Open-GGUF-Q4_1

List all available models

lemonade list

Hermes Agent

How to use alphabell/alphaBell-Qwen3.8-27B-Open-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 alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

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 alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use alphabell/alphaBell-Qwen3.8-27B-Open-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1

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 "alphabell/alphaBell-Qwen3.8-27B-Open-GGUF:Q4_1" \
  --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"

alphaBell Qwen Open 27B matched GGUF

The Q4_1 and Q8_0 text-decoder GGUF release of alphaBell Qwen Open 27B, using the exact files measured in the Model Observatory experiments.

Try the blind comparison · All five experiments · Recorded results · alphaBell

Files

File	Bytes	SHA-256
qwen-q8_0.gguf	28595763712	`b2a3172342ff220a8f9a06fc2fd252d1e84615ce8dfc00b2a66941610aa8f43c`
qwen-q4_1.gguf	16829921792	`7c0814054c69120ff0e5b3ac3f625f98a9f86185fc39f5d71eae02cb54eb5825`

Q4_1 and Q8_0 are pure quantization choices with the runtime-required floating-point tensors retained. File size is not a deployment memory requirement. See checkpoints.json for architecture hashes, tensor counts and selected GGUF metadata.

Provenance

This is alphaBell's post-trained Qwen derivative, combining abliteration and supervised fine-tuning. Both GGUF files were converted directly from the final alphaBell BF16 checkpoint, then quantized with llama-quantize --pure. They are not quantizations of the published FP8 file. Vision and the multi-token prediction head are excluded. The parent architecture revision is 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0. The related FP8 release documents training and the earlier BF16 evaluation. Those results do not establish the performance of these quantizations.

Run and reproduce

Use the pinned CPU engine source and experiment runner. The source is reconstructed from a public llama.cpp base and the included Flint-support patch; the build script verifies the resulting tree. The measured experiments use eight CPU threads, a 2048-token context, temperature 0, seed 42, reasoning disabled and a 64-token output limit. Normal model applications can use other settings, but those runs are separate measurements.

Download a file with the Hugging Face CLI:

hf download alphabell/alphaBell-Qwen3.8-27B-Open-GGUF qwen-q4_1.gguf --local-dir ./checkpoints

The files were tested with the published pinned engine. Other runtimes require compatible architecture and tokenizer support. The experiment runner checks native model dimensions against the supplied configuration before generating.

Measured comparison

The fixed study contains six prompts per model, each at Q4_1 and Q8_0. Eleven of the twelve total answer pairs were text-identical. The one differing arithmetic pair produced two incorrect answers. These short, single-generation cases establish neither general quantization parity nor a speed ranking. The downloadable records retain every prompt, answer, check and timing.

License and credit

Use and redistribution follow the included LICENSE. Model development credit remains with the source publisher and alphaBell's derivative work as described above. The viewer and experiment harness have their own code license; they do not change the checkpoint license.