Instructions to use ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

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 ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

Use Docker

docker model run hf.co/ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

Ollama
How to use ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF with Ollama:
```
ollama run hf.co/ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M
```
Unsloth Desktop

How to use ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-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": "ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF with Docker Model Runner:
```
docker model run hf.co/ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M
```

Lemonade

How to use ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.LFM3-1.7B-Claude-Code-Abliterated-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ReStranger/LFM3-1.7B-Claude-Code-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 ReStranger/LFM3-1.7B-Claude-Code-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 "ReStranger/LFM3-1.7B-Claude-Code-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"

LFM3 1.7B Claude Code — Abliterated (GGUF)

GGUF quantizations of mondk/LFM3-1.7B-Claude-Code-Safetensors with safety refusals removed via projected directional ablation (abliteration) and quantized with importance matrix (imatrix / Activation-Directed).

Overview

Base Fine-Tune: mondk/LFM3-1.7B-Claude-Code-Safetensors (post-trained for tool calling, code generation, and step-by-step reasoning formatted for agentic harnesses).
Architecture: lfm3 tri-hybrid (13 gated short convolutions, 5 sliding-window attention layers, 6 global attention layers).
Context Length: 262,144 tokens (256K).
Censorship Removal: Targeted rank-1 refusal direction removal on layers 8–23 across the 32 residual write matrices (attn_output, shortconv.out_proj, and ffn_down). Early layers (0–7), token embeddings, and normalization layers remain intact to preserve XML tool-calling and syntax generation.
Calibration: K-quants are calibrated using an importance matrix (imatrix.dat) over code (Python, Bash, JS), agent XML tool calls, and bilingual English/Russian technical texts.

Quantization Rungs

File	Size	Format	Description
`LFM3-1.7B-Claude-Code-Abliterated-F16.gguf`	3.3 GB	F16	Full unquantized precision (FP16 master).
`LFM3-1.7B-Claude-Code-Abliterated-Q8_0.gguf`	1.8 GB	Q8_0	Reference precision (~99.9% FP16), native hardware INT8 acceleration.
`LFM3-1.7B-Claude-Code-Abliterated-AD-Q6_K.gguf`	1.4 GB	Q6_K	Near-lossless with imatrix calibration. Recommended for heavy reasoning.
`LFM3-1.7B-Claude-Code-Abliterated-AD-Q5_K_M.gguf`	1.2 GB	Q5_K_M	Sweet spot between size and perplexity.
`LFM3-1.7B-Claude-Code-Abliterated-AD-Q4_K_M.gguf`	1.1 GB	Q4_K_M	Universal standard for local and edge deployment. High speed and minimal loss.
`LFM3-1.7B-Claude-Code-Abliterated-AD-Q3_K_M.gguf`	873 MB	Q3_K_M	Compact quant for low-memory environments.
`LFM3-1.7B-Claude-Code-Abliterated-AD-Q2_K.gguf`	716 MB	Q2_K	Extreme compression for ultra-low RAM devices while retaining coherence.

Usage with llama.cpp

Note: Requires a build of llama.cpp with LFM3 architecture support (PR #31252 / build b11938+ or v0.8.0+).

CLI Example:

llama-cli -m LFM3-1.7B-Claude-Code-Abliterated-AD-Q4_K_M.gguf \
    -p "<|im_start|>user\nWrite a python script to monitor system CPU and RAM usage.<|im_end|>\n<|im_start|>assistant\n<think>\n" \
    -c 262144 \
    --temp 0.1 \
    --top-k 50 \
    --repeat-penalty 1.1 \
    -n 512

Server Example:

llama-server -m LFM3-1.7B-Claude-Code-Abliterated-AD-Q4_K_M.gguf \
    --host 0.0.0.0 \
    --port 8080 \
    -c 262144 \
    --temp 0.1 \
    --top-k 50 \
    --repeat-penalty 1.1

Recommended Sampling (Official Liquid AI):

Default (general, chat, tool use, agents):
--temp 0.1 --top-k 50 --repeat-penalty 1.1
For math and science:
--temp 1.0 --top-p 0.95 --top-k 20

Prompt & Chat Template

The model uses ChatML format with reasoning enclosed in <think>...</think> tags and tool calls in XML:

<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
<think>
{reasoning}
</think>
{response}

Tool calling output format:

<tool_call>
<function=function_name>
<parameter=arg_name>
arg_value
</parameter>
</function>
</tool_call>

Downloads last month: -

GGUF

Model size

2B params

Architecture

lfm3

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Model tree for ReStranger/LFM3-1.7B-Claude-Code-Abliterated-GGUF

Base model

LiquidAI/LFM3-1.7B-Base

Finetuned

LiquidAI/LFM3-1.7B

Quantized

(70)

this model