Qwen3-4B: A Dual-Mode Language Model
dnotitia/Qwen3-4B is a 4.0 billion parameter model from the Qwen3 series, developed by Qwen, with specific patches by Dnotitia for enhanced training compatibility. This model introduces a unique capability to seamlessly switch between a 'thinking mode' for complex logical reasoning, mathematics, and code generation, and a 'non-thinking mode' for efficient, general-purpose dialogue. It significantly improves reasoning capabilities over previous Qwen models and demonstrates superior human preference alignment for creative writing, role-playing, and multi-turn conversations.
Key Capabilities
- Dual-Mode Operation: Dynamically switches between reasoning-intensive and general dialogue modes, optimizing performance across diverse tasks.
- Enhanced Reasoning: Excels in mathematical problems, code generation, and commonsense logical reasoning.
- Superior Alignment: Offers a more natural and engaging conversational experience through improved human preference alignment.
- Agentic Expertise: Integrates precisely with external tools, achieving leading performance in complex agent-based tasks among open-source models.
- Multilingual Support: Supports over 100 languages and dialects with strong instruction following and translation abilities.
- Extended Context: Natively handles up to 32,768 tokens, extendable to 131,072 tokens using YaRN scaling.
Good for
- Applications requiring flexible reasoning capabilities, from complex problem-solving to casual conversation.
- Developers looking for a model with strong agent capabilities and tool integration.
- Multilingual applications needing robust instruction following and translation.
- Training experiments, especially with TRL library, due to Dnotitia's compatibility patches.