How to use from
Docker Model Runner
# Gated model: Login with a HF token with gated access permission
hf auth login
docker model run hf.co/KissTheHabit/IDA_Edge
Quick Links

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

IDA EDGE Body

KissTheHabit/IDA_Edge is the lightweight EDGE-side artifact repository for the IDA family.

The active architecture is IDA Lattice, a custom causal language model architecture. Legacy Hub metadata may still identify older uploaded artifacts as GPT-NeoX. That metadata is not authoritative for current native Lattice training runs.

Architecture

The EDGE body preserves the same Lattice physiology as the AI body at a smaller deployment envelope:

  • recurrent selective state
  • local-attention workspace
  • controlled merging
  • cognitive-pressure routing
  • lateral inhibition
  • routed cognitive circuits
  • governed memory
  • thalamic routing
  • prefrontal workspace summaries
  • action gating
  • student-state output
  • shared GH expression projection
  • future-token auxiliary prediction
  • fidelity verification

Current EDGE Tiers

Tier Hidden Size Layers Heads Intermediate Context Purpose
edge/swift 256 6 4 768 1,024 Lightweight comparison and deployment tier
edge/full 384 6 6 1,024 Recommended production EDGE tier

Current edge/swift Proof Shape

  • Approximate parameters: ~54M
  • Recurrent state size: 192
  • Local attention window: 128
  • Workspace: 6 × 192
  • Student state size: 192
  • Future prediction horizon: 1
  • Thalamic route count: 4
  • Action gate size: 6

Paired-Body Lineage

EDGE is not a reduced generic assistant.

It is the lightweight half of a paired EDGE ↔ AI lineage. EDGE preserves the family’s identity, routing, and governed-control surfaces under a smaller local deployment budget. AI carries the deeper body for cases that justify the additional compute.

Promotion Rule

An uploaded EDGE artifact is not automatically canonical.

Promotion requires:

  • paired EDGE and AI evidence
  • identity agreement across both bodies
  • held-out evaluation
  • lineage packet completion
  • terminal promotion approval

Evidence Surfaces

Intended Use

This repository is intended for controlled edge deployment, local inference research, paired-body identity validation, and governed RegOS integration.

It is not presented as a general-purpose internet-scale assistant.

License

Business Source License 1.1, with Change Date 2028-01-01.

Apache-2.0 applies after the Change Date. Commercial use before the Change Date requires explicit permission.

Downloads last month
-
Safetensors
Model size
52.8M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support