DualMind: Dual-Mental-Modality Reasoning

DualMind, developed by Convergent Intelligence LLC, is a 1.7 billion parameter model built on the Qwen3ForCausalLM architecture. Its core innovation is dual-mental-modality reasoning, where a single model simulates an internal dialogue using specific role tokens:

• <explore>: For unconstrained reasoning and derivation.
• <examine>: For adversarial self-critique and error detection of the explore output.
• <response>: For synthesizing a clean, refined final answer.

This approach allows the model to perform self-correction, mimicking the benefits of multi-model collision arrays within a single architecture. It provides a structured mechanism for the model to make mistakes, critique them, and then refine its output, leading to more robust problem-solving.

Key Capabilities

• Structured Self-Correction: Employs a unique explore-examine-response loop for internal verification and refinement.
• Enhanced Logical Inference: Trained on the LogicInference_OA dataset, focusing on transforming complex CoT solutions into its dual-modality format.
• Qwen3ForCausalLM Base: Utilizes a DISC-refined uncensored Qwen3 base model, providing a strong foundation for reasoning tasks.
• Context Length: Features a substantial context length of 40,960 tokens.

Good For

• Complex Problem Solving: Ideal for tasks requiring deep reasoning and self-correction, such as mathematical proofs or logical puzzles.
• Robust Output Generation: Provides a mechanism to reduce errors and improve the quality of generated responses through internal critique.
• Research into Cognitive Architectures: Offers a practical implementation of dialectical reasoning within a single LLM.