reaperdoesntknow/Symiotic-14B

TEXT GENERATIONPricing:Input $0.48 / Output $0.96Concurrent Unit Cost:1Model Size:14BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 6, 2025License:afl-3.0Architecture:Transformer0.0K Featherless Exclusive Cold

reaperdoesntknow/Symiotic-14B is a 17.8-billion-parameter hybrid symbolic–transformer model based on Qwen-14B, developed by Convergent Intelligence LLC: Research Division. It integrates neural representation with structured symbolic cognition, featuring persistent memory, multi-stage symbolic routing, and self-organizing knowledge structures. This experimental model is designed for advanced reasoning agents, long-form symbolic theorem generation, and scientific dialogue in complex problem domains, supporting a 32768 token context length.

Loading preview...

Overview of SymbioticLM-14B

SymbioticLM-14B is a 17.8-billion-parameter hybrid symbolic–transformer model developed by Convergent Intelligence LLC: Research Division. Built upon the Qwen-14B backbone, it uniquely combines high-capacity neural processing with structured symbolic cognition, aiming for full-scale cognitive reasoning. This experimental research checkpoint introduces persistent memory, entropic recall, multi-stage symbolic routing, and self-organizing knowledge structures.

Key Architectural Features

  • Hybrid Design: Integrates a Qwen-14B transformer with specialized symbolic modules like ThoughtDynamicsLNN, LiquidThoughtProcessor, CrystallineProcessor (DNAConv GNN), and HelicalDNAProcessor.
  • Persistent Memory: Utilizes 4096 symbolic states (FP32) retrieved via entropy and contextual similarity, enabling true memory in conversational agents.
  • Cognitive Mechanisms: Features a "Dream Mode" for background symbolic simulation and an intelligent router for processor path selection based on intent and entropy.
  • Symbolic Tag Support: The tokenizer includes special tags like <D_LIM> and <PROOF> to facilitate symbolic processing.

Good for

  • Developing multi-step conversational agents with long-term memory capabilities.
  • Long-form symbolic theorem generation and proof planning.
  • Scientific dialogue, symbolic simulations, and complex math/code synthesis.
  • Reasoning in fuzzy, discontinuous, or non-smooth problem domains where traditional LLMs may struggle.

Limitations

  • Memory utility is dependent on curation and seeding.
  • Not instruction-tuned for general question-answering tasks.
  • Increased VRAM usage due to FlashAttention and symbolic modules.

Finetunes
1 models