Qwen3-4B-Instruct-2507-Q4_K_M-GGUF

Model Introduction

This is the GGUF format quantized version of Qwen3-4B-Instruct-2507, suitable for local deployment and inference using llama.cpp and compatible frameworks.

Original Model: Qwen/Qwen3-4B-Instruct

Quantization Information

• Quantization Method: Q4_K_M (4-bit Medium Quality)
• File Size: ~2.5 GB
• Quantization Quality: Balanced between model size and performance, suitable for most use cases

Q4_K_M uses 4-bit quantization for most weights while keeping important layers in higher precision, offering a good trade-off between speed and quality.

Usage

llama.cpp Command Line

# Interactive chat
./llama-cli -m Qwen3-4B-Instruct-2507-Q4_K_M.gguf -ngl 40 -c 4096 --chat-template qwen

# Server mode
./llama-server -m Qwen3-4B-Instruct-2507-Q4_K_M.gguf -ngl 40 -c 4096 --port 8080

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama(
    model_path="Qwen3-4B-Instruct-2507-Q4_K_M.gguf",
    n_gpu_layers=40,
    n_ctx=4096,
)

response = llm.create_chat_completion(
    messages=[{"role": "user", "content": "Hello, how are you?"}]
)
print(response["choices"][0]["message"]["content"])

Hardware Requirements

| Configuration | Minimum RAM | Recommended RAM |

|--------------|-------------|-----------------|

| CPU-only | 6 GB | 8 GB |

| GPU (NVIDIA) | 4 GB VRAM | 6 GB VRAM |

Recommended Hardware:

• GPU: NVIDIA RTX 3060 (12GB) or higher
• CPU: Modern multi-core processor with AVX2 support
• RAM: 8GB+ system memory

Download Links

• ModelScope: Voconly/Qwen3-4B-Instruct-2507-Q4_K_M-GGUF
• HuggingFace: Voconly/Qwen3-4B-Instruct-2507-Q4_K_M-GGUF

License

This model follows the original Qwen3 license. Please refer to the original model's license for usage terms.