Instructions to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"

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 webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"

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 "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Ling 3.0 Tiny Budanmış Türkçe Hukuk Modeli (Apple Silicon MLX 4-bit)

Bu depo, budanmış MoE mimarisine (bailing_hybrid) sahip ve 11 Milyonluk Türkiye Mahkeme Kararları Korpusu ile zenginleştirilmiş Ling 3.0 Tiny Budanmış Türkçe Hukuk Modelinin Apple Silicon (M1, M2, M3, M4) çipleri için optimize edilmiş 4-bit MLX sürümüdür.

🚀 Model ve Kuantalama Özellikleri

Temel Mimari: bailing_hybrid (Pruned MoE, ~1.3B aktif parametre)
Kuantalama: MLX 4-bit Affine (q_bits=4, q_group_size=64)
RAM / VRAM Gereksinimi: ~3.0 - 3.5 GB Birleşik Bellek (Unified Memory)
Hedef Donanım: 8 GB, 16 GB, 24 GB+ Apple Silicon Mac cihazlar (MacBook Air/Pro, Mac mini, Mac Studio)
Çıkarım Hızı: ÖSYM TRLawBench üzerinde soru başına 0.23 saniye ortalama çıkarım hızı (ultra hızlı!)

💻 Apple Silicon Mac'te Kullanım Rehberi

1. Kütüphaneleri Yükleyin

pip install -U mlx-lm huggingface_hub

2. Hugging Face Girişi (Model Özel/Private Olduğundan)

huggingface-cli login --token <YOUR_HF_TOKEN>

3. Python ile Hızlı Çıkarım

from mlx_lm import load, generate

model_id = "webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit"
model, tokenizer = load(model_id)

prompt = "<|im_start|>system\nSen uzman bir Türk Hukuku asistanısın.<|im_end|>\n<|im_start|>user\nTBK m. 138 aşırı ifa güçlüğünün temel şartları nelerdir?<|im_end|>\n<|im_start|>assistant\n"

response = generate(
    model,
    tokenizer,
    prompt=prompt,
    max_tokens=512,
    verbose=True
)
print(response)

4. Yerel OpenAI Uyumlu Sunucu Başlatma

python -m mlx_lm server --model webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit --port 8080

📊 Karşılaştırmalı Türk Hukuku Benchmark Sonuçları

1. Resmi ÖSYM TRLawBench Başarısı (97 Soru)

Sıra	Model	Mimari / Kuantalama	Mod	Toplam Doğru	Başarı (%)	Hız (sn/soru)	Disk Boyutu
🥇	Qwen 3.8 27B	27B Dense (`UD-Q4_K_M`)	Doğrudan (Düz)	55 / 97	%56.70	0.44 s	15.33 GB
🥈	MiniCPM-V 4.7 35B-A3B	35B MoE (`IQ4_NL` A3.6B)	Doğrudan (Düz)	51 / 97	%52.58	0.26 s	18.42 GB
🥉	Ternary-Bonsai-2-27B ⚡	27B Ternary (`PQ2_0` 2.13bpw)	Doğrudan (Düz)	49 / 97	%50.52	0.83 s	6.71 GB
4	Gemma-4-12B-it (Base) 🌟	12B Dense (`Q4_K_M`)	Doğrudan (Düz)	44 / 97	%45.36	0.76 s	6.63 GB
5	Qwen 3.5 9B (Base) 🚀	9B Dense (`Q4_K_M`)	Doğrudan (Düz)	43 / 97	%44.33	0.64 s	5.29 GB
6	Gemma-4 TR-Hukuk v2 (Pruned) 🇹🇷	20.2B MoE (`IQ4_NL` A3.2B)	Doğrudan (Düz)	38 / 97	%39.18	0.23 s	10.92 GB
7	Ling-3.0-tiny (Base) ⚡	7.9B MoE (`Q4_K_M` A1.3B)	Doğrudan (Düz)	31 / 97	%31.96	0.77 s	4.49 GB
8	ling3-tr-tiny-pruned-hukuk 🇹🇷	Pruned MoE (MLX 4-bit & GGUF)	Doğrudan (Düz)	29 / 97	%29.90	0.23 s	2.84 GB / 3.5 GB
9	Ternary-Bonsai-2-9B ⚡	9B Ternary (`PQ2_0` 2.13bpw)	Doğrudan (Düz)	29 / 97	%29.90	0.38 s	2.25 GB

2. Türk Hukuku Mevzuat & Terminoloji Hakimiyeti (Held-out 750 Soru Cloze Testi)

Model	Taban (Pre-train)	Fine-Tuned (Post-train)	Net İlerleme / Artış
Qwen 3.5 9B Türkçe Hukuk 🚀	%41.87 (314 / 750)	%47.47 (356 / 750)	+5.60 puan (+42 Soru)
Gemma-4 26B-A4B Pruned Hukuk v2	%27.60 (207 / 750)	%30.40 (228 / 750)	+2.80 puan (+21 Soru)

⚖️ Lisans

Özel Mülkiyet (proprietary-private). Yetkisiz dağıtımı yasaktır.

Downloads last month: 7

Safetensors

Model size

5B params

Tensor type

BF16

F32

U32

MLX

Hardware compatibility

4-bit

Model tree for webbrain-one/ling3-tr-tiny-pruned-hukuk-MLX-4bit

Base model

webbrain-one/ling3-turkish-tiny-pruned

Quantized

(1)

this model