Tuwhy/Qwen3-1.7B-OPSA
Tuwhy/Qwen3-1.7B-OPSA is a 1.7 billion parameter causal language model from the Qwen3 series, enhanced with On-Policy Self-Adaptation (OPSA) for improved performance. This model excels particularly in mathematical reasoning and complex problem-solving tasks, demonstrating significant gains over its base counterpart on benchmarks like AIME and HMMT. It supports a 32,768 token context length and can operate in both thinking and non-thinking modes, offering flexibility for various applications.
Loading preview...
Qwen3-1.7B-OPSA: Enhanced Reasoning with On-Policy Self-Adaptation
Tuwhy/Qwen3-1.7B-OPSA is a 1.7 billion parameter model built upon the Qwen3 architecture, specifically fine-tuned using the On-Policy Self-Adaptation (OPSA) method. OPSA is a novel training approach that improves model policy without relying on a teacher model, reward model, or external task rewards, focusing on self-improvement.
Key Capabilities and Differentiators
- Superior Mathematical Reasoning: The OPSA training significantly boosts the model's performance on challenging math benchmarks. For instance, it achieves 48.85 / 80.00 on AIME24 and 23.33 / 50.00 on HMMT25, marking substantial improvements over the base Qwen3-1.7B.
- Flexible Thinking Modes: Inheriting from the Qwen3 series, this model supports seamless switching between a 'thinking mode' for complex logical reasoning, math, and coding, and a 'non-thinking mode' for efficient, general-purpose dialogue. The OPSA training and evaluation were conducted in non-thinking mode.
- Agentic Capabilities: Qwen3 models, including this variant, demonstrate strong tool-calling abilities, recommended for use with the Qwen-Agent framework for complex agent-based tasks.
- Optimized for Efficiency: With 1.7 billion parameters and a 32,768 token context length, it offers a balance of capability and efficiency, making it suitable for applications requiring strong reasoning in a smaller footprint.
When to Use This Model
This model is particularly well-suited for:
- Mathematical Problem Solving: Its enhanced reasoning makes it ideal for applications requiring accurate mathematical computations and step-by-step problem-solving.
- Complex Reasoning Tasks: Scenarios demanding logical deduction and structured thought processes can benefit from its OPSA-driven improvements.
- Resource-Constrained Environments: As a 1.7B parameter model, it offers strong performance without the computational overhead of much larger models.
For optimal results, users should adhere to the recommended sampling parameters and ensure enable_thinking=False in the chat template when aiming to reproduce the reported OPSA performance.