reaperdoesntknow/DualMind
DualMind by Convergent Intelligence LLC is a 1.7 billion parameter Qwen3ForCausalLM model designed for dual-mental-modality reasoning. It employs role tokens (, , ) to simulate internal self-critique and refinement within a single architecture, leveraging a 40,960 token context length. This model excels at complex logical inference and problem-solving by structurally enabling self-correction and dialectical reasoning.
Loading preview...
DualMind: Dual-Mental-Modality Reasoning
DualMind is a 1.7 billion parameter model from Convergent Intelligence LLC that introduces a unique dual-mental-modality reasoning approach. It operates with a single architecture but simulates two internal "voices" or modalities, differentiated by special role tokens: <explore>, <examine>, and <response>. This allows the model to perform unconstrained reasoning, critically self-evaluate its own output, and then synthesize a refined final answer.
Key Capabilities
- Structured Self-Correction: The model's internal dialogue mechanism (explore, examine, response) provides a structural method for self-correction, allowing it to make mistakes in the explore phase and then identify and refine them in the examine phase.
- Dialectical Reasoning: Recreates the dynamic of multi-model collision arrays within a single architecture, where different perspectives lead to novel insights.
- Logical Inference: Specifically trained on logical inference problems from the KK04/LogicInference_OA dataset, formatted to leverage its dual-modality structure.
- Qwen3ForCausalLM Architecture: Built upon a DISC-refined uncensored Qwen3 base model, featuring 2.03B parameters (1.7B effective) and a substantial 40,960 token context length.
Good For
- Complex Problem Solving: Ideal for tasks requiring deep reasoning, where a single-pass approach might miss nuances or introduce errors.
- Verification and Refinement: Useful in scenarios where output accuracy and robustness are critical, as the model actively critiques its own derivations.
- Research into Cognitive Architectures: Demonstrates a novel method for integrating self-correction and multi-perspective reasoning into a single LLM.