Instructions to use mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf mertkayacs/Karar-4B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/mertkayacs/Karar-4B-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use mertkayacs/Karar-4B-GGUF with Ollama:
```
ollama run hf.co/mertkayacs/Karar-4B-GGUF:Q4_K_M
```
Unsloth Desktop

How to use mertkayacs/Karar-4B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mertkayacs/Karar-4B-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": "mertkayacs/Karar-4B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use mertkayacs/Karar-4B-GGUF with Docker Model Runner:
```
docker model run hf.co/mertkayacs/Karar-4B-GGUF:Q4_K_M
```

Lemonade

How to use mertkayacs/Karar-4B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull mertkayacs/Karar-4B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Karar-4B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-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 mertkayacs/Karar-4B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use mertkayacs/Karar-4B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mertkayacs/Karar-4B-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 "mertkayacs/Karar-4B-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"

Karar-4B GGUF: small Turkish language model for local decisions

This project is by Eschatia Labs, an independent non-profit open-source AI research lab.

GGUF builds of Karar-4B, a small Turkish language model for text classification and decisions on a CPU. Full-precision Karar-4B: 96.8% accuracy; Kev-4B: 87.1% on held-out Turkish decisions from the training data pipeline. This is a full-precision comparison; GGUF agreement is measured separately below.

Quick facts

Item	Value
Quantized from	mertkayacs/Karar-4B
Q4_K_M file size	2.71 GB
Q5_K_M file size	3.08 GB
Q8_0 file size	4.48 GB
Peak RAM (Q4_K_M, 4k context)	3.03 GB
Same selected option (Q4_K_M)	96.7% vs full-precision
Hugging Face downloads	672 (as of 2026-10-15)
GitHub stars	5

Project links

Cite this

@software{kaya2026jevalt,
  author = {Kaya, Mert},
  title  = {JevAlt: Small Decision Models for English, Turkish and German},
  year   = {2026},
  license = {Apache-2.0},
  url    = {https://github.com/mertkayacs/jevalt}
}

Run with JevAlt and llama.cpp

Q4_K_M is the default, with 3.03 GB peak RAM at a 4k context.

pip install "jevalt[serve,gguf] @ git+https://github.com/mertkayacs/jevalt" && jevalt serve --model mertkayacs/Karar-4B-GGUF --file Karar-4B-Q4_K_M.gguf

Send requests to http://127.0.0.1:8000/v1/systemone. The main card has a complete request. The JevAlt server reads next-token decision probabilities through llama-cpp-python and applies calibration.json.

In Ollama: ollama run mertkayacs/karar-4b (library page) answers with the option letter.

Plain llama.cpp, Ollama and LM Studio chat endpoints return generated text; use the JevAlt server for the Jev API and calibrated decisions.

Files and quantization checks

File	Size	Selected-option agreement	Maximum probability gap	Seconds per request	Peak RAM	Load setting
Q4_K_M	2.71 GB	96.7%	0.1430	15.74	3.03 GB	no mmap, repack on
Q5_K_M	3.08 GB	100.0%	0.1154	28.17	3.39 GB	no mmap, repack on
Q8_0	4.48 GB	100.0%	0.0396	18.56	4.77 GB	no mmap, repack on

Selected-option agreement compares each quantization with full-precision Karar-4B on 61 validation decisions per quantization. It is a quantization check rather than a benchmark score. The Q4_K_M build selects the same option on 96.7% of those decisions. More precision does not guarantee higher agreement on this small sample.

export-report.json records probability gaps, speed and RAM. RAM was measured in a fresh process on an 8-vCPU HF machine, 4 threads, 4k context, with no mmap and repacking enabled.

Use and limits

Use for routing, tagging and triage in Turkish. Refit calibration on your own data before setting thresholds. Planted instructions, long irrelevant text and date arithmetic remain failure cases; keep authorization outside the model.

The widget records a full-precision answer. Full results and training | Try the Space | Code | Project page.

Apache-2.0. Quantized from mertkayacs/Karar-4B.

An Eschatia Labs project. Mert Kaya.