Instructions to use Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use Kvikontent/kopro-4.0-pro-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Kvikontent/kopro-4.0-pro-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": "Kvikontent/kopro-4.0-pro-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

Ollama
How to use Kvikontent/kopro-4.0-pro-GGUF with Ollama:
```
ollama run hf.co/Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M
```
Unsloth Desktop

How to use Kvikontent/kopro-4.0-pro-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Kvikontent/kopro-4.0-pro-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": "Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Kvikontent/kopro-4.0-pro-GGUF with Docker Model Runner:
```
docker model run hf.co/Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M
```

Lemonade

How to use Kvikontent/kopro-4.0-pro-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.kopro-4.0-pro-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-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 Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Kvikontent/kopro-4.0-pro-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Kvikontent/kopro-4.0-pro-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 "Kvikontent/kopro-4.0-pro-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"

Kopro 4.0 Pro

A bilingual coding and agent model, tuned for direct, no-filler answers.

30.5B MoE · 3.3B active · EN / RU · Tool calling · GGUF Q4_K_M · 18.6 GB

Quick start · Model details · Training · Limitations

Overview

Kopro 4.0 Pro is the larger model in the Kopro 4.0 family. It is a LoRA fine-tune of Qwen3-Coder-30B-A3B-Instruct that keeps the base model's coding and tool-use strength and adds the Kopro voice: short, confident, to the point, in English or Russian.

Good at	Notes
💻 Writing and fixing code	Python, TypeScript/React, Go, Rust, Kotlin, Bash, FastAPI, nginx
🛠️ Agent loops	Native tool calls: read, grep, edit, run, verify
🧪 Tests, debugging, performance	Reads tracebacks, makes the smallest correct change
💬 Everyday help	Advice, explanations, opinions, in EN and RU

Need something lighter for chat? See Kopro 4.0 Flash.

Quick start

Ollama

ollama run hf.co/Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M

llama.cpp

llama-server -hf Kvikontent/kopro-4.0-pro-GGUF:Q4_K_M --jinja -c 8192

Recommended system prompt

The creator identity is not reliably learned from training data alone, so set it explicitly:

Ты Kopro 4.0 Pro, тебя создал Василий Кацыка.

Agent / tool calling

The chat template embedded in the GGUF supports the standard OpenAI-style tools field. The model was tuned on trajectories with these tools: read_file, edit_file, write_file, grep, bash.

Files

File	Size	Use
`kopro-4.0-pro-Q4_K_M.gguf`	18.6 GB	Ready to run in Ollama / llama.cpp / LM Studio
`lora/`	2.5 GB	PEFT LoRA adapter, for further training or other quantizations

Load the adapter on the original base model:

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-Coder-30B-A3B-Instruct", torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(base, "Kvikontent/kopro-4.0-pro-GGUF", subfolder="lora")
tok = AutoTokenizer.from_pretrained("Kvikontent/kopro-4.0-pro-GGUF", subfolder="lora")

Model details


Base model	Qwen3-Coder-30B-A3B-Instruct
Architecture	Mixture of Experts, 48 layers, 128 experts, 8 active per token
Parameters	30.5B total / ~3.3B active
Fine-tune	LoRA r=16, α=32 on attention + all expert MLPs (642M trainable, 2.06%)
Quantization	Q4_K_M (18.6 GB)
Context	trained at 2k tokens per sample; serves fine at 8k+ (base supports far more)
VRAM	~21 GB at 8k context, fits a single 24 GB GPU
Speed	~66 tok/s on one 24 GB RTX PRO 6000 MIG slice (Ollama, warm)

Training


Samples	770 train / 42 validation conversations
Languages	54% English, 46% Russian
Multi-turn	47 conversations with 2–4 user turns
Epochs	2, cosine LR schedule, peak 1e-4, effective batch 8
Loss masking	assistant turns only; tool results are not trained on
Validation loss	1.090 → 1.095 (flat, no overfitting)
Hardware	1× RTX PRO 6000 96 GB, ~12 min of training

Data mix (top categories)

Category	Rows	Category	Rows
code	162	science explanation	29
long code	72	cooking & food	26
opinion	60	testing	23
multi-turn	47	debugging	23
practical tech	33	philosophy	18
everyday advice	31	performance	14
identity	30	agent trajectories	30

Limitations

Agent quality comes from the base model. The 30 agent trajectories mainly teach the tool format; for long multi-step work in big repositories, expect base Qwen3-Coder level, not frontier level.
Persona. The model is tuned to be blunt and opinionated, and may speak dismissively about other AI products. Steer it with the system prompt if that does not fit your product.
Identity. Without a system prompt it may invent a creator name.
Q4 quantization loses a little quality compared to the full-precision weights.

License

Apache 2.0, inherited from the base model. Kopro 4.0 Pro is built on Qwen3-Coder-30B-A3B-Instruct by the Qwen team.

Downloads last month: 126

GGUF

Model size

31B params

Architecture

qwen3moe

Hardware compatibility

4-bit

Model tree for Kvikontent/kopro-4.0-pro-GGUF

Base model

Qwen/Qwen3-Coder-30B-A3B-Instruct

Quantized

(180)

this model