Text Generation
Transformers
GGUF
PEFT
Russian
qwen
qwen3.5
lora
russian
instruct
unsloth
alpaca
conversational
Instructions to use LLiserginov/Luqwen-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use LLiserginov/Luqwen-4B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="LLiserginov/Luqwen-4B") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("LLiserginov/Luqwen-4B", device_map="auto") - PEFT
How to use LLiserginov/Luqwen-4B with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use LLiserginov/Luqwen-4B 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 LLiserginov/Luqwen-4B:Q4_K_M # Run inference directly in the terminal: llama cli -hf LLiserginov/Luqwen-4B:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf LLiserginov/Luqwen-4B:Q4_K_M # Run inference directly in the terminal: llama cli -hf LLiserginov/Luqwen-4B: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 LLiserginov/Luqwen-4B:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf LLiserginov/Luqwen-4B: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 LLiserginov/Luqwen-4B:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf LLiserginov/Luqwen-4B:Q4_K_M
Use Docker
docker model run hf.co/LLiserginov/Luqwen-4B:Q4_K_M
- LM Studio
- Jan
- vLLM
How to use LLiserginov/Luqwen-4B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "LLiserginov/Luqwen-4B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "LLiserginov/Luqwen-4B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/LLiserginov/Luqwen-4B:Q4_K_M
- SGLang
How to use LLiserginov/Luqwen-4B with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "LLiserginov/Luqwen-4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "LLiserginov/Luqwen-4B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "LLiserginov/Luqwen-4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "LLiserginov/Luqwen-4B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Ollama
How to use LLiserginov/Luqwen-4B with Ollama:
ollama run hf.co/LLiserginov/Luqwen-4B:Q4_K_M
- Unsloth Studio
How to use LLiserginov/Luqwen-4B with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for LLiserginov/Luqwen-4B to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for LLiserginov/Luqwen-4B to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for LLiserginov/Luqwen-4B to start chatting
- Pi
How to use LLiserginov/Luqwen-4B with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf LLiserginov/Luqwen-4B:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @mariozechner/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "LLiserginov/Luqwen-4B:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use LLiserginov/Luqwen-4B with Docker Model Runner:
docker model run hf.co/LLiserginov/Luqwen-4B:Q4_K_M
- Lemonade
How to use LLiserginov/Luqwen-4B with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull LLiserginov/Luqwen-4B:Q4_K_M
Run and chat with the model
lemonade run user.Luqwen-4B-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use LLiserginov/Luqwen-4B with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf LLiserginov/Luqwen-4B: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 LLiserginov/Luqwen-4B:Q4_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use LLiserginov/Luqwen-4B with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf LLiserginov/Luqwen-4B: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 "LLiserginov/Luqwen-4B: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"
| language: ru | |
| license: cc-by-nc-4.0 | |
| library_name: transformers | |
| base_model: Qwen/Qwen3.5-4B-Base | |
| tags: | |
| - qwen | |
| - qwen3.5 | |
| - lora | |
| - peft | |
| - russian | |
| - instruct | |
| - unsloth | |
| - alpaca | |
| pipeline_tag: text-generation | |
| datasets: | |
| - LLiserginov/russian-instructions-10k | |
| widget: | |
| - messages: | |
| - role: user | |
| content: "Напиши короткий рассказ о коте и луне." | |
| # Luqwen-4B | |
| **Luqwen-4B** — русскоязычная текстовая модель, полученная дообучением **Qwen3.5-4B-Base** методом **LoRA (QLoRA)** на наборе русскоязычных инструкций. Модель предназначена для следования инструкциям на русском языке. | |
| Базовая модель Qwen3.5-4B-Base использует гибридную архитектуру (Gated DeltaNet + Full Attention) с контекстом до 262k токенов, что обеспечивает эффективный инференс при сохранении качества генерации. | |
| ## Base model | |
| | Свойство | Значение | | |
| |---|---| | |
| | Архитектура | Qwen3.5 For Causal LM (гибрид Linear/Full Attention) | | |
| | Параметров | 4B | | |
| | Скрытая размерность | 1024 | | |
| | Слоёв | 24 | | |
| | Контекст | до 262 144 токенов | | |
| | Вокабуляр | 248 320 (padding) | | |
| | MTP | 1 слой | | |
| Подробнее: [Qwen3.5-4B-Base](https://huggingface.co/Qwen/Qwen3.5-4B-Base) | |
| ## Training data | |
| Модель дообучалась на датасете [russian-instructions-10k](https://huggingface.co/datasets/LLiserginov/russian-instructions-10k) — ~10k русскоязычных пар инструкция-ответ (CC BY-NC 4.0). | |
| **Pipeline подготовки данных:** | |
| 1. Фильтрация 10k примеров из [Alpaca Cleaned](https://github.com/gururise/AlpacaDataCleaned) (все coding + math, остальные random) | |
| 2. Перевод с английского на русский через **Gemma 4 26B** (llama.cpp API) | |
| 3. Очистка от непереведённых примеров (27 записей) | |
| 4. Конвертация в ChatML-формат | |
| **Состав датасета:** | |
| | Категория | Количество | | |
| |---|---| | |
| | General | ~8 812 | | |
| | Math | ~603 | | |
| | Coding | ~560 | | |
| | **Total** | **~9 975** | | |
| ## Training procedure | |
| ### Параметры LoRA | |
| | Параметр | Значение | | |
| |---|---| | |
| | rank (r) | 16 | | |
| | lora_alpha | 32 | | |
| | lora_dropout | 0 | | |
| | bias | none | | |
| | Целевые модули | `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`, `in_proj_qkv`, `in_proj_z`, `out_proj`, `in_proj_a`, `in_proj_b` | | |
| ### Гиперпараметры обучения | |
| | Параметр | Значение | | |
| |---|---| | |
| | Оптимизатор | AdamW 8-bit | | |
| | Precision | BF16 (mixed) | | |
| | Batch size | 1 (8 gradient accumulation steps) | | |
| | Learning rate | 2e-4 | | |
| | Эпохи | 3 | | |
| | Warmup steps | 20 | | |
| | Max seq length | 4096 токенов | | |
| | Обёртка | unsloth (4-bit QLoRA) | | |
| ### Слияние весов | |
| После обучения LoRA-адаптер был слит с базовой моделью в 16-bit точность (метод `merged_16bit` через unsloth) для удобного использования без дополнительных зависимостей PEFT. | |
| ## Usage | |
| ```python | |
| from transformers import AutoModelForCausalLM, AutoTokenizer | |
| model_name = "LLiserginov/Luqwen-4B" | |
| tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) | |
| model = AutoModelForCausalLM.from_pretrained( | |
| model_name, | |
| torch_dtype="auto", | |
| device_map="auto", | |
| trust_remote_code=True, | |
| ) | |
| messages = [ | |
| {"role": "user", "content": "Напиши короткий рассказ о коте и луне."}, | |
| ] | |
| text = tokenizer.apply_chat_template( | |
| messages, | |
| tokenize=False, | |
| add_generation_prompt=True, | |
| ) | |
| inputs = tokenizer(text, return_tensors="pt").to(model.device) | |
| outputs = model.generate( | |
| **inputs, | |
| max_new_tokens=1024, | |
| temperature=0.7, | |
| top_p=0.9, | |
| do_sample=True, | |
| ) | |
| response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True) | |
| print(response) | |
| ``` | |
| ### Использование с vLLM | |
| ```python | |
| from vllm import LLM, SamplingParams | |
| llm = LLM(model="LLiserginov/Luqwen-4B", trust_remote_code=True) | |
| messages = [ | |
| {"role": "user", "content": "Объясни разницу между supervised и unsupervised learning."}, | |
| ] | |
| outputs = llm.chat(messages, sampling_params=SamplingParams(temperature=0.7, max_tokens=1024)) | |
| print(outputs[0].outputs[0].text) | |
| ``` | |
| ## Limitations | |
| - Модель дообучена на **малом объёме данных** (~10k примеров), что может ограничивать качество и разнообразие ответов | |
| - Датасет переведён машинным способом (Gemma 4 26B) — возможны артефакты, буквализмы и потеря смысла | |
| - Модель не проходила RLHF/DPO-калибровку и может генерировать нежелательный или фактически неверный контент | |
| - Не предназначена для использования в медицинских, юридических или других критических областях | |
| - Это текстовая версия — мультимодальные возможности Qwen3.5 (изображения, видео) не используются | |
| ## License | |
| Модель распространяется под лицензией **CC BY-NC 4.0** (Creative Commons Attribution Non Commercial 4.0) ввиду ограничений производного датасета. | |
| Базовая модель: [Qwen3.5-4B-Base](https://huggingface.co/Qwen/Qwen3.5-4B-Base) — Apache 2.0. | |