---
license: apache-2.0
language:
- en
- multilingual
datasets:
- agentlans/LinguaNova
base_model:
- agentlans/Qwen3-4B-multilingual-sft
tags:
  - qwen
  - multilingual
  - quantized
  - llama.cpp
  - natural-language-processing
  - text-generation
  - efficient-inference
  - lightweight-model
---
# Qwen 3 4B Multilingual Quantized Models

This repo contains quantized versions of the [agentlans/Qwen3-4B-multilingual-sft](https://huggingface.co/agentlans/Qwen3-4B-multilingual-sft) model, optimized for efficient local use with **llama.cpp**.

The models were quantized using an [unofficial Docker image](https://hub.docker.com/r/agentlans/llamacpp) and calibrated on the first 100 rows of the [LinguaNova dataset](https://huggingface.co/datasets/agentlans/LinguaNova) to maintain strong multilingual performance.

These quantized models share the same strengths and limitations as the original Qwen 3 4B multilingual model. They offer a lighter, faster alternative for inference with minor trade-offs in precision.