reaperdoesntknow/Symiotic-14B

SymbioticLM-14B: Hybrid Symbolic–Transformer Model

SymbioticLM-14B is a 17.8 billion parameter model developed by reaperdoesntknow, combining a Qwen-14B transformer backbone with advanced symbolic processing. This hybrid architecture aims to surpass traditional LLMs in symbolic domains by tightly coupling neural and symbolic cognition.

Key Capabilities

• Hybrid Architecture: Integrates a Qwen-14B transformer with specialized symbolic modules like ThoughtDynamicsLNN, LiquidThoughtProcessor, CrystallineProcessor (DNAConv GNN), and HelicalDNAProcessor.
• Persistent Memory: Features 4096 symbolic states in FP32, retrieved using entropy and contextual similarity, enabling true memory for multi-step interactions.
• Advanced Reasoning: Supports multi-stage symbolic routing and self-organizing knowledge structures, grounded in Discrepancy Calculus for dynamic completeness and stability.
• Dream Mode: Includes a background symbolic simulation for open-ended cognitive processes.
• Symbolic Tag Support: Utilizes a custom tokenizer with added tokens for symbolic tags like <D_LIM> and <PROOF>.

Good For

• Advanced Reasoning Agents: Ideal for complex cognitive tasks requiring deep understanding and structured thought.
• Symbolic Math & Code Generation: Excels in long-form symbolic theorem generation, proof planning, and synthesizing mathematical or programming constructs.
• Scientific Dialogue & Simulation: Suitable for scientific discourse and symbolic simulations in fuzzy or discontinuous problem domains.
• Conversational Agents with Memory: Designed for multi-step conversations that require true, persistent memory and entropic recall.