Model Overview

Just-Bax/Qwen3-14B-Base-Uzbek-Cyrillic is a 14.8 billion parameter causal language model, fine-tuned from the Qwen/Qwen3-14B-Base architecture. This model leverages a Transformer Decoder architecture and was fine-tuned using LoRA (Low-Rank Adaptation) with Unsloth, a training framework designed for efficient fine-tuning. It processes inputs with bfloat16 precision and supports a substantial context length of 32,768 tokens.

Key Capabilities

• Uzbek Cyrillic Text Generation: Excels at producing natural and grammatically correct text in the Uzbek Cyrillic script.
• Multilingual Support: Preserves the inherent multilingual capabilities of its base Qwen3-14B model, making it suitable for broader language applications.
• Efficient Fine-tuning: Utilizes LoRA with specific parameters (r=16, α=32, dropout=0.0) for targeted adaptation.

Intended Use Cases

• Content Generation: Ideal for creating various forms of text content in Uzbek Cyrillic.
• Conversational AI: Suitable for developing chatbots and dialogue systems in Uzbek.
• Text Summarization: Can be used for summarizing documents and articles in Uzbek Cyrillic.
• Central Asian Language Applications: Applicable in multilingual setups that involve Central Asian languages, particularly Uzbek.