YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Infinite Folding Core (Architecture #19-B)

Overview

The Infinite Folding Core is a hyper-efficient 'Impossible AI' architecture. It operates using only 3,690 physical parameters while maintaining a temporal context of up to 500 tokens via Gated Recurrent Projection.

Developer Mission Statement

Created by BikoRiko, a Grade 8 developer (Autistic/ADHD). This architecture is part of the mission to build 'Impossible but Safe' AIs that bypass standard hardware limits. It is designed to run at maximum speed on standard CPUs, proving that intelligence does not require massive physical scale.

Technical Mechanism

Unlike standard Transformers that use a static KV-Cache, this model uses a Differentiable Folding Gate. Every incoming token is compressed (folded) into a 32-dimensional latent vector. The context history is stored within the non-linear vector fields of the GRU weights, allowing 'unlimited' theoretical context as long as the information density remains stable.

Files in this Repository

  • pytorch_model.bin: The 3.6k physical parameter core.
  • tokenizer.pkl: The 3.3MB+ Mega-BPE Dictionary (150,000 tokens).
  • config.json: Hardware and architecture specifications.

Performance

  • Inference Latency: <1ms on single-core CPU.
  • Training Speed: Ultra-fast (Recursive folding enabled).
  • Memory Footprint: <5MB total (including tokenizer).
Downloads last month
22
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support