Instructions to use LiquidAI/LFM2.5-VL-3B-DSpark-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 LiquidAI/LFM2.5-VL-3B-DSpark-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 LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
# Run inference directly in the terminal:
llama cli -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
# Run inference directly in the terminal:
llama cli -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

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 LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
# Run inference directly in the terminal:
./llama-cli -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

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 LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

Use Docker

docker model run hf.co/LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

LM Studio
Jan

vLLM

How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "LiquidAI/LFM2.5-VL-3B-DSpark-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": "LiquidAI/LFM2.5-VL-3B-DSpark-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/LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

Ollama
How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with Ollama:
```
ollama run hf.co/LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
```
Unsloth Desktop

How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

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": "LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with Docker Model Runner:
```
docker model run hf.co/LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16
```

Lemonade

How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

Run and chat with the model

lemonade run user.LFM2.5-VL-3B-DSpark-GGUF-F16

List all available models

lemonade list

Hermes Agent

How to use LiquidAI/LFM2.5-VL-3B-DSpark-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 LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

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 LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use LiquidAI/LFM2.5-VL-3B-DSpark-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16

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 "LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16" \
  --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"

Try LFM • Docs • Discord

LFM2.5-VL-3B-DSpark-GGUF

GGUF build of LiquidAI/LFM2.5-VL-3B-DSpark for llama.cpp.

This is a standalone draft sidecar: it carries only the drafter (4 attention layers, a Markov head, a confidence head, and block size 9.) Token embeddings and the LM head are shared from the target model at load time, so it must be paired with a LFM2.5-VL-3B-GGUF target file.

Find more information about LFM2.5-VL-DSpark in our blog post.

📦 Files

file	size	notes
`LFM2.5-VL-3B-DSpark-F16.gguf`	567 MB	Standalone F16 drafter; pair with the target model and vision projector from `LiquidAI/LFM2.5-VL-3B-GGUF`.

All inference numbers for this release use 16-bit processing for both the vision encoder and language backbone.

🏃 How to run (llama.cpp)

Run the target model with the DSpark drafter:

llama-server -hf LiquidAI/LFM2.5-VL-3B-GGUF:F16 \
  -hfd LiquidAI/LFM2.5-VL-3B-DSpark-GGUF:F16 \
  --spec-type draft-dspark --spec-draft-n-max 8 --spec-draft-n-min 0 \
  -ngl 99 -ngld 99 -fa on

The drafter was trained with block size 9. For Apple silicon, we recommend block size 8 through --spec-draft-n-max 8.

Speculative decoding is exact under greedy decoding: the target verifies every proposed token, so the generated output equals the target model running alone. The llama.cpp timing logs report the draft acceptance rate.

Other LFM2.5-VL-3B-DSpark formats:

Draft	Target
LFM2.5-VL-3B-DSpark	LFM2.5-VL-3B
LFM2.5-VL-3B-DSpark-GGUF	LFM2.5-VL-3B-GGUF

📊 Acceptance and benchmarks

See LiquidAI/LFM2.5-VL-3B-DSpark for model details and acceptance-length and throughput benchmarks across six vision-language tasks on NVIDIA H100 and Apple silicon, including MLX-VLM, llama.cpp, and SGLang results.

📬 Contact

Got questions or want to connect? Join our Discord community
If you are interested in custom solutions with edge deployment, please contact our sales team.

Citation

@article{liquidAI2026VL3B,
  author  = {Liquid AI},
  title   = {LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge},
  journal = {Liquid AI Blog},
  year    = {2026},
  note    = {www.liquid.ai/blog/lfm2-5-vl-3b},
}

@article{liquidAI2026vldspark,
  author = {Liquid AI},
  title = {LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond},
  journal = {Liquid AI Blog},
  year = {2026},
  note = {www.liquid.ai/blog/lfm2-5-vl-dspark},
}