Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled
Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled is a 9 billion parameter reasoning model, fine-tuned on the Qwen3.5-9B dense architecture. It leverages Chain-of-Thought (CoT) distillation from Claude-4.6 Opus interactions, excelling in structured step-by-step problem-solving within a 32768 token context window. This model is optimized for breaking down complex problems, planning methodologies using internal tags, and delivering precise solutions, making it ideal for analytical tasks, coding, and mathematics.
Loading preview...
Model Overview
Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled is a 9 billion parameter model built upon the Qwen3.5-9B architecture, specifically fine-tuned for advanced reasoning capabilities. It incorporates Chain-of-Thought (CoT) distillation from Claude-4.6 Opus interactions, focusing on structured problem-solving. The model has been further enhanced with additional reasoning data, including the Jackrong/Qwen3.5-reasoning-700x dataset, to improve structured step-by-step reasoning and diversity.
Key Capabilities
- Structured Thinking: Employs a streamlined reasoning paradigm, planning steps within
<think>tags to reduce redundant cognitive loops and improve inference efficiency. - High-Density Reasoning Logic: Achieved through Supervised Fine-Tuning (SFT) using Unsloth, with loss calculated purely over the generation of
<think>sequences and subsequent solutions. - Extended Context Support: Supports a 16,384 token context window, allowing for complex multi-step reasoning traces.
- Distilled Performance: Training loss converged from 0.5138 to 0.35786, indicating successful internalization of structured reasoning patterns from Claude 4.6 Opus teacher data.
Intended Use Cases
- Analytical Tasks: Best suited for offline analytical tasks requiring transparent, step-by-step logic.
- Coding & Mathematics: Excels in heavy logic-dependent prompting, such as code generation and mathematical problem-solving.
- Problem Decomposition: Ideal for scenarios where users need to follow the AI's internal logic in breaking down and solving complex problems.