reaperdoesntknow/Symbiotic-1B

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 6, 2025License:afl-3.0Architecture:Transformer0.0K Featherless Exclusive Warm

reaperdoesntknow/Symbiotic-1B is a 0.8 billion parameter hybrid symbolic-transformer model, built upon the Qwen-1B architecture by Convergent Intelligence LLC: Research Division. It integrates a rotary transformer with a symbolic processing pipeline and persistent episodic memory, enabling advanced reasoning capabilities. Optimized for CPU and embedded inference, this model excels at symbolic reasoning, procedural planning, and mathematical modeling in resource-constrained environments. Its unique architecture, based on Discrepancy Calculus, allows for efficient logical processing and memory-augmented tasks.

Loading preview...

Symbiotic-1B: A Hybrid Symbolic-Transformer for Efficient Reasoning

Symbiotic-1B, developed by Convergent Intelligence LLC: Research Division, is a compact 0.8 billion parameter model that uniquely combines a Qwen-1B rotary transformer backbone with a sophisticated symbolic processing pipeline and persistent episodic memory. This architecture is designed for lightweight, memory-augmented reasoning, making it suitable for CPU and embedded inference.

Key Capabilities & Architecture Highlights

  • Hybrid Design: Fuses a Qwen-1B transformer with symbolic modules like ThoughtDynamicsLNN, CrystallineProcessor (DNAConv GNN), LiquidThoughtProcessor, and HelicalDNAProcessor.
  • Memory-Augmented: Features 2048 symbolic vectors with entropic and contextual retrieval for enhanced reasoning.
  • Dream Mode: Includes a symbolic simulation capability with ThoughtGenerator for complex queries.
  • Discrepancy Calculus Foundation: Developed under the DISC framework, which treats training singularities as structural signals for learning geometry.

Ideal Use Cases

  • CPU-optimized symbolic inference in constrained environments.
  • Educational agents requiring memory and logical processing.
  • Generating graph-based explanations.
  • Procedural planning, mathematical modeling, and small-scale code generation.

Limitations

  • Less fluent in free-form language compared to larger models.
  • Symbolic accuracy improves with careful memory curation.
  • Dream Mode may require warm-up or symbolic seeding for intricate tasks.

Finetunes
1 models