DavidAU/Qwen3.5-2B-Polaris-HighIQ-Thinking-Compact
DavidAU/Qwen3.5-2B-Polaris-HighIQ-Thinking-Compact is a 2.3 billion parameter Qwen3.5-based language model fine-tuned by DavidAU on the Polaris 1000x dataset. This model is optimized for efficient reasoning, producing terse outputs and directly generating responses when 'thinking' is not required. It features a 32768 token context length and demonstrates improved performance over the base Qwen3.5-2B model on various benchmarks, particularly in reasoning and instruction following tasks.
Loading preview...
Model Overview
DavidAU/Qwen3.5-2B-Polaris-HighIQ-Thinking-Compact is a 2.3 billion parameter model built on the Qwen3.5 architecture, fine-tuned by DavidAU using the Polaris 1000x dataset. It is designed for efficient processing, often providing concise outputs and directly generating responses without extensive 'thinking' traces when appropriate. The model supports a substantial context length of 32,768 tokens, with the underlying Qwen3.5 base model natively supporting 262,144 tokens.
Key Capabilities
- Optimized Reasoning: The model is specifically trained to produce very short thinking traces and terse outputs, or to generate directly when complex reasoning is not needed.
- Enhanced Performance: Benchmarks indicate superior performance compared to the base Qwen3.5-2B model across various tasks, including ARC, HellaSwag, and PIQA.
- Multimodal Foundation: Inherits Qwen3.5's unified vision-language foundation, supporting multimodal tokens for reasoning, coding, agents, and visual understanding.
- Efficient Architecture: Utilizes Gated Delta Networks and sparse Mixture-of-Experts for high-throughput inference with low latency.
- Multilingual Support: The base Qwen3.5 model expands support to 201 languages and dialects.
- Thinking Mode: Configurable to operate in a 'thinking mode' via API parameters, allowing for more deliberate processing for complex tasks.
Good For
- Prototyping and Research: Ideal for rapid development and experimental purposes due to its compact size and specialized tuning.
- Task-Specific Fine-tuning: Suitable as a base for further fine-tuning on particular applications requiring efficient, focused responses.
- Reasoning Tasks: Excels in scenarios where concise and accurate reasoning is paramount, as evidenced by its benchmark improvements.
- Multimodal Applications: Can be leveraged for applications requiring both text and visual understanding, such as image and video analysis, inherited from the Qwen3.5 foundation.