Instructions to use daveycodez/LFM3-1.7B-Web-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 daveycodez/LFM3-1.7B-Web-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 daveycodez/LFM3-1.7B-Web-GGUF
# Run inference directly in the terminal:
llama cli -hf daveycodez/LFM3-1.7B-Web-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf daveycodez/LFM3-1.7B-Web-GGUF
# Run inference directly in the terminal:
llama cli -hf daveycodez/LFM3-1.7B-Web-GGUF

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf daveycodez/LFM3-1.7B-Web-GGUF
# Run inference directly in the terminal:
./build/bin/llama cli -hf daveycodez/LFM3-1.7B-Web-GGUF

Use Docker

docker model run hf.co/daveycodez/LFM3-1.7B-Web-GGUF

Llama new
LM Studio
Jan

vLLM

How to use daveycodez/LFM3-1.7B-Web-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "daveycodez/LFM3-1.7B-Web-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": "daveycodez/LFM3-1.7B-Web-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/daveycodez/LFM3-1.7B-Web-GGUF

Ollama
How to use daveycodez/LFM3-1.7B-Web-GGUF with Ollama:
```
ollama run hf.co/daveycodez/LFM3-1.7B-Web-GGUF
```
Unsloth Desktop

How to use daveycodez/LFM3-1.7B-Web-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf daveycodez/LFM3-1.7B-Web-GGUF

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": "daveycodez/LFM3-1.7B-Web-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use daveycodez/LFM3-1.7B-Web-GGUF with Docker Model Runner:
```
docker model run hf.co/daveycodez/LFM3-1.7B-Web-GGUF
```

Lemonade

How to use daveycodez/LFM3-1.7B-Web-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull daveycodez/LFM3-1.7B-Web-GGUF

Run and chat with the model

lemonade run user.LFM3-1.7B-Web-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use daveycodez/LFM3-1.7B-Web-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 daveycodez/LFM3-1.7B-Web-GGUF

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 daveycodez/LFM3-1.7B-Web-GGUF

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use daveycodez/LFM3-1.7B-Web-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf daveycodez/LFM3-1.7B-Web-GGUF

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 "daveycodez/LFM3-1.7B-Web-GGUF" \
  --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"

LFM3-1.7B GGUF for the web

A smaller 4-bit GGUF of LFM3-1.7B by Liquid AI, made for in-browser use with wllama (llama.cpp in WebAssembly) on the CPU, where phones without WebGPU run it. It's the GGUF counterpart of daveycodez/LFM3-1.7B-Web-ONNX and uses the same vocabulary.

This is an unofficial community build. It is not made, reviewed or endorsed by Liquid AI.

Changes from the original

This repository contains modified files. Compared with LiquidAI/LFM3-1.7B-GGUF:

Smaller vocabulary: trimmed from 128,000 to 49,012 BPE tokens (plus the same 140 special tokens; 49,152 embedding rows), the same vocabulary as daveycodez/LFM3-1.7B-Web-ONNX. Every byte token and all special tokens (chat, tool-call and thinking tokens) are kept, and the merges stay consistent, so any text still tokenizes and decodes back exactly. The tied embedding / LM head, the tokenizer and the special token ids changed to match. Less common words split into more tokens, and non-English text needs many more tokens (see below).
Re-quantized weights: quantized from the BF16 GGUF to Q4_K with an importance matrix, with the tied embedding / LM head at Q6_K (90.7% of the file Q4_K, 9.1% Q6_K). The importance matrix came from about 36k tokens of WikiText-2 training text plus chat replies generated by the BF16 model, and covers the output layer too. No calibration text is included in the file.

The chat template and sampling settings in the file are unchanged.

	Q4_K_M (original)	This build
File	1,099,627,808 bytes	911,948,864 bytes (−188 MB)
WikiText-2 test perplexity, per original token (BF16: 19.79)	20.36	20.15
Perplexity on chat replies written by the trimmed BF16 model (BF16: 2.639)	2.773	2.770
Perplexity on chat replies written by the original BF16 model (BF16: 2.615)	2.746	3.260
Prompts answered with a direct tool call (BF16: 20/20)	20/20	20/20

Perplexity is scored per original token so the two vocabularies compare fairly. On replies written by the original model, this build scores worse mostly on the words the smaller vocabulary splits, which the model never saw split in training; on its own replies it matches the original, since it picks words that tokenize well.

Tokens: English prompts take about 1–2% more tokens than with the original vocabulary. Other languages take far more: German about 1.75×, Spanish 1.6×, Japanese 2.5×, Chinese 3×, Russian 2.8× and Arabic 4.2× as many, which makes prompts slower to read and replies slower to write. For non-English use, prefer the original.

Speed: in headless Chromium on a Mac with wllama at 4 threads, it loads about 27% faster and writes about 6% faster than the original Q4_K_M; prompt reading is the same.

Usage

Use it like the original GGUF, with a llama.cpp build that supports LFM3 (commit 19cce69 or later), in the browser with wllama or natively. Thinking is switched with the chat template's enable_thinking.

License

LFM3 is released under the LFM Open License v1.0, which also covers this build. Commercial use is licensed only for entities with annual revenue under $10 million. Original model © Liquid AI, Inc.

Downloads last month: 77

GGUF

Model size

2B params

Architecture

lfm3

Hardware compatibility

Model tree for daveycodez/LFM3-1.7B-Web-GGUF

Base model

LiquidAI/LFM3-1.7B-Base

Finetuned

LiquidAI/LFM3-1.7B

Quantized

(79)

this model