reaperdoesntknow/Dualmind-Qwen-1.7B-Thinking

Dualmind-Qwen-1.7B-Thinking: Deliberative Reasoning at 1.7B Parameters

This model, developed by Convergent Intelligence LLC: Research Division, is a 1.7 billion parameter Qwen3-based language model specifically trained to emulate the deliberative reasoning patterns of Claude Opus 4.6. Utilizing the DualMind SFT methodology, it was fine-tuned on over 2.5 million tokens from the Opus-4.6-Reasoning-3000x-filtered dataset. Unlike models trained on synthetic logic, Dualmind-Qwen-1.7B-Thinking learns to navigate uncertainty, backtrack, hedge, and synthesize information, reflecting a more genuine cognitive process.

Key Capabilities & Features

• Opus-like Reasoning: Absorbs the "shape of deliberation" from Claude Opus 4.6, including self-correction and nuanced synthesis.
• DualMind SFT Methodology: Leverages a specialized Supervised Fine-Tuning approach to distill complex reasoning traces into a smaller model.
• Robust Base: Built upon Disctil-Qwen3-1.7B, a DISC-refined model, providing a strong foundational structure.
• Extended Context: Supports a maximum context length of 40,960 tokens, allowing for long-form reasoning and generation.
• BF16 Precision: Trained and available in BF16 for efficient inference.

Ideal Use Cases

• Complex Problem Solving: Suited for tasks requiring multi-step reasoning, analysis, and nuanced decision-making.
• Cognitive Simulation: Useful for applications where emulating human-like deliberation and self-correction is beneficial.
• Research in Reasoning: Provides a compact model for exploring the distillation of advanced reasoning capabilities from larger, frontier models.
• Long-form Generation: Its training on extended reasoning chains makes it capable of producing longer, more coherent outputs, especially when paired with appropriate generation parameters (e.g., higher max_new_tokens, temperature 0.6-0.8, repetition_penalty 1.1-1.2).