Instructions to use madebywest/HorizonAI-230M-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 madebywest/HorizonAI-230M-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 madebywest/HorizonAI-230M-GGUF:Q4_0
# Run inference directly in the terminal:
llama cli -hf madebywest/HorizonAI-230M-GGUF:Q4_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf madebywest/HorizonAI-230M-GGUF:Q4_0
# Run inference directly in the terminal:
llama cli -hf madebywest/HorizonAI-230M-GGUF:Q4_0

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 madebywest/HorizonAI-230M-GGUF:Q4_0
# Run inference directly in the terminal:
./llama-cli -hf madebywest/HorizonAI-230M-GGUF:Q4_0

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 madebywest/HorizonAI-230M-GGUF:Q4_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf madebywest/HorizonAI-230M-GGUF:Q4_0

Use Docker

docker model run hf.co/madebywest/HorizonAI-230M-GGUF:Q4_0

LM Studio
Jan
Ollama
How to use madebywest/HorizonAI-230M-GGUF with Ollama:
```
ollama run hf.co/madebywest/HorizonAI-230M-GGUF:Q4_0
```
Unsloth Desktop

How to use madebywest/HorizonAI-230M-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf madebywest/HorizonAI-230M-GGUF:Q4_0

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": "madebywest/HorizonAI-230M-GGUF:Q4_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use madebywest/HorizonAI-230M-GGUF with Docker Model Runner:
```
docker model run hf.co/madebywest/HorizonAI-230M-GGUF:Q4_0
```

Lemonade

How to use madebywest/HorizonAI-230M-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull madebywest/HorizonAI-230M-GGUF:Q4_0

Run and chat with the model

lemonade run user.HorizonAI-230M-GGUF-Q4_0

List all available models

lemonade list

Hermes Agent

How to use madebywest/HorizonAI-230M-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 madebywest/HorizonAI-230M-GGUF:Q4_0

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 madebywest/HorizonAI-230M-GGUF:Q4_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use madebywest/HorizonAI-230M-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf madebywest/HorizonAI-230M-GGUF:Q4_0

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 "madebywest/HorizonAI-230M-GGUF:Q4_0" \
  --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"

HorizonAI-230M · QAD-Q4_0 GGUF

HorizonAI-230M is a full-parameter SFT + QAT fine-tune of LiquidAI/LFM2.5-230M, trained on-device (Apple Silicon, MLX) for multilingual instruction following.

This repository ships the QAD-style Q4_0 build: the model was trained with a bit-exact straight-through-estimator simulation of llama.cpp Q4_0 round-to-nearest quantization (blocks of 32, d = signed_extremum / -8, half-up rounding, fp16 scales), so the quantized weights you run are exactly what the model was optimized for. Embeddings are quantized to Q6_K (mirroring LiquidAI's official QAD layout). No importance-matrix / calibration is involved.

File

File	Quant	Size
`HorizonAI-230M-QAD-Q4_0.gguf`	Q4_0 (+ Q6_K embedding)	142 MB

Training

Base: LiquidAI/LFM2.5-230M (BF16)
Data: ~73k instruction samples — smol-constraints, Tulu-3 persona-IF, smol-rewrite, smol-magpie-ultra, OpenHermes, APIGen function calling, MetaMathQA, everyday conversations, plus Italian (ita_chat, ~10k) and language-balanced multilingual data (it/es/pt/fr/de quotas) to prevent language drift
Recipe: full fine-tune (bf16 master, AdamW, warmup+cosine), stage 1 SFT, stage 2 QAT (70%) with fake-quant active; loss on assistant turns only, tokenization aligned with inference-time rendering
Hardware: Apple M4 (16 GB), MLX, ~5 h

Results (measured, same harness for every model)

32 machine-checked instruction-following prompts (EN+IT, incl. 8 language-fidelity checks) at temp 0.1 / top-k 50 / rep-penalty 1.05; perplexity on wikitext-2 test (llama-perplexity, 200 chunks, ctx 2048).

Model	Size	IF-score	Languages (it/fr/es/pt/de)	PPL
HorizonAI-230M-QAD-Q4_0	142 MB	40.6%	5/5	35.6
LFM2.5-230M Q6_K (stock)	182 MB	40.6%	4/5	47.9
LFM2.5-230M Q6_K official	182 MB	37.5%	4/5	47.9
LFM2.5-230M QAD-Q4_0 official	142 MB	28.1%	3/5	45.0

Chinese/Japanese/Korean remain at base-model level (no targeted SFT data).

Usage

llama-cli -m HorizonAI-230M-QAD-Q4_0.gguf --chat-template ...

Runs in llama.cpp, LM Studio, and any GGUF runtime with lfm2 architecture support.

License & attribution

This model is a derivative work of LiquidAI/LFM2.5-230M and is redistributed under the LFM Open License v1.0 (see NOTICE). The fine-tuning was performed by Dylan Succi for the HorizonAI / TravelApp project. Commercial use is permitted under the license's revenue threshold (< $10M annual revenue).

Downloads last month: 36

GGUF

Model size

0.2B params

Architecture

lfm2

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for madebywest/HorizonAI-230M-GGUF

Base model

LiquidAI/LFM2.5-230M-Base

Finetuned

LiquidAI/LFM2.5-230M

Finetuned

(42)

this model