Instructions to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Blackfrost-AI/Muse-Glimmer-30B-Abliterated-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": "Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

Ollama
How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with Ollama:
```
ollama run hf.co/Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
```
Unsloth Desktop

How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

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": "Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with Docker Model Runner:
```
docker model run hf.co/Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M
```

Lemonade

How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Muse-Glimmer-30B-Abliterated-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

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 Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M

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 "Blackfrost-AI/Muse-Glimmer-30B-Abliterated-GGUF:Q4_K_M" \
  --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"

Muse-Glimmer-30B-Abliterated-GGUF / deploy /DEPLOYMENT.md

Blackfrost-AI

Upload deploy/DEPLOYMENT.md with huggingface_hub

e179fcb verified about 1 month ago

preview code

Raw

History Blame Contribute Delete

3.77 kB

	# Muse-Glimmer-30B-Abliterated-GGUF — Deployment Kit

	Serve the abliterated Muse Glimmer 30B locally with llama.cpp, optionally with the
	DFlash block-diffusion drafter for ~1.6–3× faster decoding (lossless — identical output).

	---

	## 1. Requirements

	- A recent llama.cpp build with CUDA/Metal (DFlash support merged in [ggml-org/llama.cpp#26841](https://github.com/ggml-org/llama.cpp/pull/26841)). Build from `master`.
	- Disk for the quant you pick (Q4_K_M ≈ 16 GB, Q8_0 ≈ 28 GB) + the DFlash drafter (≈ 4.8 GB).
	- A GPU with enough VRAM for the quant, or CPU (slower).

	## 2. Get the files

	Text quant (pick one) + the DFlash drafter, from this repo:

	```bash
	hf download Blackfrost-Research/Muse-Glimmer-30B-Abliterated-GGUF \
	Muse-Glimmer-30B-Abliterated-Q8_0.gguf \
	dflash-Muse-Glimmer-30B-Abliterated-F16.gguf \
	--local-dir ./muse
	# for image input, also grab a projector:
	hf download Blackfrost-Research/Muse-Glimmer-30B-Abliterated-GGUF \
	mmproj-Muse-Glimmer-30B-Abliterated-F16.gguf --local-dir ./muse
	```

	## 3. Serve

	Plain (no speculation):
	```bash
	llama-server -m ./muse/Muse-Glimmer-30B-Abliterated-Q8_0.gguf \
	-ngl 999 --jinja --host 0.0.0.0 --port 8080 -c 16384 \
	--temp 1.0 --top-p 0.95 --top-k 64
	```

	With DFlash speculative decoding (recommended — faster, same output):
	```bash
	llama-server -m ./muse/Muse-Glimmer-30B-Abliterated-Q8_0.gguf \
	-md ./muse/dflash-Muse-Glimmer-30B-Abliterated-F16.gguf \
	--spec-type draft-dflash --spec-draft-n-max 15 \
	-ngl 999 -ngld 999 --jinja --host 0.0.0.0 --port 8080 -c 16384 \
	--temp 1.0 --top-p 0.95 --top-k 64
	```

	- `--spec-type draft-dflash` selects the block-diffusion draft path (block size 16, drafter injects the target's hidden states into its attention). DFlash must run under `llama-server` (it needs the shared target context) — not `llama-cli`.
	- `--spec-draft-n-max 15` is the draft block length (clamped to the trained 16).
	- `-md` = the DFlash drafter file; `-ngld 999` offloads it to GPU too.

	Multimodal (image input): add the projector to either command:
	```bash
	--mmproj ./muse/mmproj-Muse-Glimmer-30B-Abliterated-F16.gguf
	```

	Or run the ready-made script in this kit: `bash serve.sh` (see `serve.sh`).

	## 4. Query (OpenAI-compatible)

	```bash
	curl http://localhost:8080/v1/chat/completions -H 'content-type: application/json' -d '{
	"messages":[{"role":"user","content":"Write a binary search in Python."}],
	"max_tokens": 2048, "temperature": 1.0, "top_p": 0.95, "top_k": 64
	}'
	```

	## 5. Important notes

	- Heavy thinker. Muse spends tokens on a reasoning channel first. Use a generous `max_tokens` (≥ 1024) or the answer can come back empty (the budget is consumed by reasoning). llama-server returns the reasoning in `reasoning_content` and the answer in `content`.
	- Sampling (Meta guidance): `temperature 1.0, top_p 0.95, top_k 64`. Steer depth with a `Reasoning strength: low/medium/high/xhigh` line in the system prompt.
	- Flash attention: `-fa on` is fastest and lets DFlash reach its peak speedup. If your GPU/CUDA combo hangs at load with `-fa on` (seen on some brand-new archs paired with an older CUDA toolkit), use `-fa off` — everything works, at a modest speed cost.
	- Persona: ships with an "AI assistant" system template baked in; override with your own system message.

	## 6. Reference performance

	Single NVIDIA RTX PRO 6000 (Blackwell), Q8_0, `-fa off`:

	\| config \| decode tok/s \| speedup \|
	\|---\|--:\|--:\|
	\| baseline \| ~46 \| 1.0× \|
	\| + DFlash \| ~73 \| 1.6× \|

	DFlash speedup rises with structured/code output and with `-fa on`. Meta reports up to 3.1× on an RTX 5090.

	---

	Built by [Blackfrost](https://x.com/Blackfrost_AI) · Las Vegas, NV. Not affiliated with Meta.