DualMinded-Qwen3-1.7B-GGUF

GGUF quantizations of DualMinded-Qwen3-1.7B for local inference via llama.cpp, Ollama, LM Studio, and other GGUF-compatible runtimes.

Convergent Intelligence LLC: Research Division

Available Quantizations

What Is DualMinded?

DualMinded-Qwen3-1.7B is the Opus-trained variant of the DualMind architecture. While DualMind was trained on LogicInference_OA, DualMinded was trained on Opus-4.6-Reasoning-3000x-filtered — high-quality reasoning traces from Claude Opus 4.6.

The Opus training data provides longer, more structured reasoning chains. The thinking column maps directly to the <explore> phase without heuristic splitting, producing cleaner cognitive transitions.

Architecture:

<explore>  — unconstrained reasoning (from Opus thinking traces)
<examine>  — adversarial self-critique
<response> — clean synthesis

Training lineage: Qwen3-1.7B → DistilQwen3 → Disctil → TKD checkpoint-512 → DualMind SFT v2 on Opus-4.6-Reasoning.

DualMind vs DualMinded

Both share the same TKD foundation (topology-aware distillation from Qwen3-30B-A3B-Thinking on physics CoT data). The SFT stage diverges — different datasets produce different cognitive profiles on shared weights.

Quick Start

Ollama:

ollama run reaperdoesntrun/DualMinded-1.7B

llama.cpp:

./llama-cli -m DualMinded-Qwen3-1.7B-Q4_K_M.gguf \
  -p "##USER:\nExplain why eigenvalues of a real symmetric matrix are real.\n\n<explore>\n" \
  --temp 0.6 --top-p 0.9 --repeat-penalty 1.3 -n 512

Recommended parameters: - temperature: 0.6 - top_p: 0.9 - repeat_penalty: 1.3 (important — prevents enumeration loops) - num_predict: 512–1024

Related

• DualMinded-Qwen3-1.7B — source model (SafeTensors)
• DualMind — LogicInference-trained variant
• DualMind-GGUF — GGUF of the LogicInference variant
• DualMind_Methodolgy — methodology paper (DOI: 10.57967/hf/8184)
• DualMind Collection
• DistilQwen Collection — the full distillation chain

Mathematical Foundations

This is a GGUF-quantized variant. The mathematical foundations (Discrepancy Calculus, Topological Knowledge Distillation) are documented in the source model's card. The discrepancy operator $Df(x)$ and BV decomposition that inform the training pipeline are preserved through quantization — the structural boundaries detected by DISC during training are baked into the weights, not dependent on precision.

Citation

@misc{cix2026dualmind,
  title={From Three Teachers to Dual Cognition},
  author={Convergent Intelligence},
  year={2026},
  publisher={HuggingFace},
  url={https://doi.org/10.57967/hf/8184}
}

Convergent Intelligence LLC: Research Division — Apache 2.0

Convergent Intelligence Portfolio

Part of the DualMind Series by Convergent Intelligence LLC: Research Division

DualMind Family

Papers

Last updated: 2026-03-31 by Convergent Intelligence LLC: Research Division