olegselajev241/qwen3 repository overview

⁠Qwen3 Large Language Models

This repository contains quantized Qwen3 models optimized for use with Docker Model runner. Qwen3 is a state-of-the-art large language model series developed by Alibaba Cloud's Qwen team.

⁠Available Tags

• 32B-Q6_K (~26.9GB) - Qwen3 32B model with Q6_K quantization
• 14B-Q6_K (~12.1GB) - Qwen3 14B model with Q6_K quantization

Both models use Q6_K quantization which provides an excellent balance between model quality and resource efficiency.

⁠Usage with Docker Model runner

# Pull and run the 14B model
docker run --rm -it -p 11434:11434 olegselajev241/qwen3:14B-Q6_K

# Pull and run the 32B model (requires more resources)
docker run --rm -it -p 11434:11434 olegselajev241/qwen3:32B-Q6_K

⁠Model Information

• Model Family: Qwen3
• Quantization: Q6_K (6-bit quantization with K-means)
• Use Cases: Text generation, question answering, code generation, multilingual tasks
• Languages: Primarily Chinese and English, with support for multiple languages

⁠Original Model Sources

• Qwen3-32B: Qwen/Qwen2.5-32B-Instruct⁠
• Qwen3-14B: Qwen/Qwen2.5-14B-Instruct⁠
• Model Collection: Qwen HuggingFace Organization⁠

⁠Requirements

• 14B Model: Minimum 16GB RAM recommended
• 32B Model: Minimum 32GB RAM recommended
• Docker with Model runner support

⁠Performance Notes

The Q6_K quantization maintains ~99% of the original model quality while significantly reducing memory requirements and improving inference speed.

These models are quantized versions of the original Qwen models. Please refer to the original HuggingFace repositories for model licenses and detailed documentation.