Instructions to use LiquidAI/LFM2.5-VL-3B-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-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-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf LiquidAI/LFM2.5-VL-3B-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 LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf LiquidAI/LFM2.5-VL-3B-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 LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf LiquidAI/LFM2.5-VL-3B-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 LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use LiquidAI/LFM2.5-VL-3B-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-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-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-GGUF:Q4_K_M

Ollama
How to use LiquidAI/LFM2.5-VL-3B-GGUF with Ollama:
```
ollama run hf.co/LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M
```
Unsloth Desktop

How to use LiquidAI/LFM2.5-VL-3B-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-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": "LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use LiquidAI/LFM2.5-VL-3B-GGUF with Docker Model Runner:
```
docker model run hf.co/LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M
```

Lemonade

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

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.LFM2.5-VL-3B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use LiquidAI/LFM2.5-VL-3B-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-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 LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use LiquidAI/LFM2.5-VL-3B-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-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 "LiquidAI/LFM2.5-VL-3B-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"

Is the actual context length 32768?

by MuAlphaOmegaEpsilon - opened 28 days ago

Discussion

MuAlphaOmegaEpsilon

28 days ago

I tried to launch > llama-server LFM2.5-VL-3B-Q4_0.gguf -c 131072just for testing and it reported the following:

0.02.592.680 W llama_context: n_ctx_seq (131072) > n_ctx_train (128000) -- possible training context overflow
0.03.071.738 W srv    load_model: the slot context (131072) exceeds the training context of the model (128000) - capping

Are you sure the context length is limited to 32,768 as reported in the Base model card?
I'm trying to figure out if this model can be used in place of the standard LFM2.5-2.6B with the additional vision support on top or if I have to make some compromises on the context length.

MuAlphaOmegaEpsilon changed discussion title from Is the actual context length to Is the actual context length 32768? 28 days ago

ImSamhel

18 days ago

+1 Would love to know the same

samuelstevens

Liquid AI org 17 days ago

•

edited 17 days ago

32,768 is the maximum context length seen during vision training. Different phases during text-only pre-training see longer contexts; but the model has never seen examples with an image and a longer context. So you should run it with -c 32768, and we'll try to fix the metadata in the GGUFs.

I'm trying to figure out if this model can be used in place of the standard LFM2.5-2.6B with the additional vision support on top or if I have to make some compromises on the context length.

Unfortunately, I think LFM2.5-VL-3B is not nearly as agentic as the 2.6B. We will try to improve the VL models' agentic capabilities in future releases, but if you absolutely need multimodal, agentic capabilities on edge right now, I've seen some demos that use the VL model to describe/perform OCR on images, then pass the input to the 2.6B. Might be worth hacking around and seeing if it works.

If you have particular use cases that you would like to address, please let me know! You can find my email on my website.

ImSamhel

15 days ago

Thank you for the comment on this <3
I'll be experimenting later how much I can push the 2.6B model for agentic codebase discovery, and general web-scraping and document-managing tasks, among other things and experiments regarding context management - considering my custom harness in the mix. VL combination as a separately hosted rest listener will probably be the route I'll go then, but a multimodal version of 2.5 would always be welcome <3

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment