🦬 Zubr Mini 1.9 3B

⚡ Development Notice: There is a new model zubr-mini-1.9-8b

Repository: alekringtonnn-ai/zubr-mini-1.9-3b

---

📖 What This Model Is

zubr-mini-1.9-3b is a multimodal (Vision) language model with 3 billion parameters, fine-tuned from Mistral 3 3B (Ministral-3-3B-Instruct-2512 by Mistral AI). Developed by alekringtonnn-ai.

The model is distributed in GGUF format — a universal format for quantized weights designed for local inference. It runs on regular PCs (CPU), laptops, mobile devices, and servers without expensive GPUs.

💡 For Macs with Apple Silicon (M1–M4), there is a separate MLX version of this same model: alekringtonnn-ai/zubr-mini-1.9-3b-mlx.

Key Features

• Multimodal (Vision): the model supports image recognition and Image-Text-to-Text tasks (image + text → text).
• Multilingual: language support expanded from 26 to 48 languages (including Russian).
• Performance: improved vocabulary, high inference speed, and deep instruction tuning.
• Local execution: the GGUF format runs on CPU and GPU via Ollama, llama.cpp, vLLM, LM Studio, and other runtimes.
• License: Apache 2.0 — free to use, including for commercial purposes.

Model Tree

mistralai/Ministral-3-3B-Base-2512
        │ (quantization / instruct version)
mistralai/Ministral-3-3B-Instruct-2512
        │
mistralai/Ministral-3-3B-Instruct-2512-GGUF
        │ (fine-tune by alekringtonnn-ai)
alekringtonnn-ai/zubr-mini-1.9-3b          ← GGUF version (THIS repository)
        │ (conversion to MLX)
alekringtonnn-ai/zubr-mini-1.9-3b-mlx       ← MLX version for Apple Silicon

---

📊 Versions (Quantizations) in the Repository

The repository offers 8 GGUF quantization options — from an ultra-compact 2-bit one to a nearly original 8-bit one:

💡 How to choose: the standard recommendation is Q4_K_M (quality/size balance). If memory is tight — Q3_K_L or IQ3_XS; if you need maximum quality — Q6_K or Q8_0.

---

📥 How to Download

Requirements

• A CPU with AVX/AVX2 support (almost any modern x86) or an ARM device; a GPU is optional.
• RAM: at least twice the size of the quantization file (see the table above).
• Free disk space: 1.5–3.7 GB depending on the version.
• For vLLM — an NVIDIA GPU (or Apple Silicon) and Linux/macOS.

Method 1: Ollama (the easiest)

# Install Ollama: https://ollama.com
# The recommended version:
ollama run hf.co/alekringtonnn-ai/zubr-mini-1.9-3b:Q4_K_M

# Other quantizations — just change the tag:
ollama run hf.co/alekringtonnn-ai/zubr-mini-1.9-3b:Q2_K
ollama run hf.co/alekringtonnn-ai/zubr-mini-1.9-3b:Q8_0

Ollama will automatically download the required GGUF file and start a chat in your terminal.

Method 2: llama.cpp

# The recommended version — download and run in one command:
llama-cli -hf alekringtonnn-ai/zubr-mini-1.9-3b:Q4_K_M

# Or download the file and specify it explicitly:
llama-cli -m zubr-mini-1.9.Q4_K_M.gguf -p "Hello! Tell me about yourself."

Method 3: huggingface-cli (direct file downloads)

# Install the CLI
pip install -U "huggingface_hub[cli]"

# Download a specific quantization:
huggingface-cli download alekringtonnn-ai/zubr-mini-1.9-3b \
  zubr-mini-1.9.Q4_K_M.gguf

# Download the entire repository (all 8 quantizations, ~17 GB):
huggingface-cli download alekringtonnn-ai/zubr-mini-1.9-3b

By default, files are saved to ~/.cache/huggingface/hub/.

Method 4: Direct links (browser / wget / curl)

curl -L -o zubr-mini-1.9.Q4_K_M.gguf \
  "https://huggingface.co/alekringtonnn-ai/zubr-mini-1.9-3b/resolve/main/zubr-mini-1.9.Q4_K_M.gguf"

The same links also work in a browser: open the file page and click "Download", or append ?download=true to the URL.

Method 5: LM Studio / Jan (no command line)

1. Install LM Studio or Jan.
2. In the search tab, type zubr-mini-1.9-3b (or paste the repository link).
3. Select a quantization and click Download.
4. Start chatting in the built-in chat.

Method 6: Docker

docker model run hf.co/alekringtonnn-ai/zubr-mini-1.9-3b:Q4_K_M

---

🚀 How to Run

Via vLLM (API server)

pip install vllm
vllm serve "alekringtonnn-ai/zubr-mini-1.9-3b"

Once started, the model is available through an OpenAI-compatible API at http://localhost:8000/v1.

Via Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="alekringtonnn-ai/zubr-mini-1.9-3b",
    filename="*Q4_K_M.gguf",  # Or any other quantization
    verbose=False,
)

response = llm.create_chat_completion(
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello! Write a short poem about a bison."},
    ]
)
print(response["choices"][0]["message"]["content"])

Via Ollama in Python

import ollama

response = ollama.chat(
    model="hf.co/alekringtonnn-ai/zubr-mini-1.9-3b:Q4_K_M",
    messages=[
        {"role": "user", "content": "Hello! Tell me about yourself."}
    ]
)
print(response["message"]["content"])

---

⚙️ Recommended Generation Parameters

---