Recursive Language Models designed By (WithIn Us AI) at core. (Experimental gated stages) Finalized versions coming soon. • 9 items • Updated • 1
👑 Royal.Opaque.Reasoner.IX
ROR-IX — Sovereign Opaque Reasoning System
“The deepest cognition occurs beyond visibility.”
⸻
🌌 Overview
Royal.Opaque.Reasoner.IX (ROR-IX) is an experimental recursive reasoning architecture developed by WithinUsAI focused on latent cognition, recursive abstraction, sovereign reasoning orchestration, and deep internal inference systems.
ROR-IX unifies multiple cognitive subsystems into a single synchronized forward-pass architecture designed to simulate reflective reasoning rather than static token prediction.
Unlike conventional language models, ROR-IX investigates:
- recursive cognition loops
- hidden-state planning
- adaptive reasoning pathways
- self-corrective inference
- latent abstraction systems
- multimodal cognitive fusion
The architecture is built around the concept that:
Intelligence is not merely output generation — it is structured internal reasoning.
⸻
👑 Identity
Royal Opaque Reasoner
The “Royal” designation represents:
- sovereign orchestration
- hierarchical cognition
- adaptive reasoning authority
- recursive oversight systems
The “Opaque” designation symbolizes:
- hidden cognition layers
- latent reasoning structures
- abstract internal planning
- compressed thought synthesis
ROR-IX is designed as:
- a recursive reasoning engine
- an experimental cognition framework
- a sovereign inference system
- a frontier AI research architecture
⸻
⚡ Model Highlights
Attribute Value Parameters ~4.897B Context Length 444,000 Tokens Precision bfloat16 Architecture Recursive Hybrid-Mind Transformer Reasoning System Multi-Expert Recursive Routing Memory System Differentiable Hybrid Memory Multimodal Support Image / Audio / Video Projection RLHF Support PPO-Compatible Value Head
⸻
🧠 Hybrid-Mind Components
All cognitive systems execute during every forward pass.
The architecture is designed to simulate synchronized recursive cognition across multiple reasoning pathways.
⸻
🔁 MetaLearningModulator
Fast-weight hypernetwork enabling dynamic adaptation and inner-loop contextual learning.
⚖️ RLValueHead
Token-level value estimation architecture for:
- PPO optimization
- RLHF workflows
- alignment experimentation
- reinforcement-guided reasoning
🧬 AdaptiveLayerNorm
Context-conditioned normalization system supporting continual adaptation and dynamic representation scaling.
🧠 ReasoningRouter
4-expert soft-routing cognition architecture specializing across:
- natural language reasoning
- logical inference
- spatial cognition
- numerical abstraction
🔮 SelfRewritingSignal
Gradient-free self-correction mechanism that recursively evaluates generation quality and reasoning consistency.
⚡ InnovationHead
Four divergent entropy-weighted attention streams designed to expand exploratory cognition and creative reasoning pathways.
🛰️ DebugProbe
Internal cognitive probes estimating:
- coherence
- contradiction
- novelty
- confidence stability
🧩 HybridMemoryBank
512-slot differentiable memory system combining:
- short-term cognition
- persistent latent memory
- contextual retrieval pathways
🌌 RecursiveSeed
256-dimensional recursive latent seed unrolled through a 3-stage GRU reflective cognition cycle.
🎥 MultiModalProjectors
Projection systems for integrating:
- image embeddings
- audio embeddings
- video embeddings
into unified hidden-state cognition space.
⸻
⚙️ Technical Specifications
Vocabulary Size : 65,536 Context Length : 444,000 Tokens Hidden Size : 2048 Layers : 32 Attention Heads : 32 KV Heads : 8 (GQA) FFN Dimension : 8192 SwiGLU RoPE Theta : 500000.0 Precision : bfloat16
⸻
💻 Fine-Tuning
Standard Causal Language Modeling
out = model(input_ids=ids, labels=ids) loss = out["loss"]
⸻
RLHF / PPO Value Optimization
out = model(input_ids=ids, return_value=True) values = out["value"] # (B, T)
⸻
🌌 Research Philosophy
ROR-IX explores the hypothesis that:
Advanced reasoning systems require recursive internal cognition.
The architecture investigates:
- reflective inference loops
- latent abstraction systems
- recursive planning architectures
- sovereign reasoning structures
- multimodal cognition fusion
- synthetic recursive intelligence
The model emphasizes:
- structured reasoning
- adaptive cognition
- hidden-state planning
- recursive refinement
- frontier-scale experimentation
⸻
⚠️ Experimental Status
Royal.Opaque.Reasoner.IX is an experimental open research model.
Human verification is recommended for:
- legal guidance
- medical information
- financial decisions
- safety-critical applications
⸻
🌵 Origin
Created by WithinUsAI Built from Albuquerque, New Mexico.
Independent frontier AI research exploring:
- recursive intelligence
- sovereign cognition systems
- latent reasoning architectures
- synthetic abstraction
- evolving AI systems
⸻ 👑 Final Motto
“The deepest reasoning remains unseen.”
License & Usage Terms
© 2026 Within Us AI. All Rights Reserved.
Protected Works
This repository contains Recursive Language Models (including all variants, weights, parameters, fine-tunes, and derivatives) and associated datasets. All materials are the exclusive intellectual property of Within Us AI.
License Summary
- All rights reserved.
- Strict internal use only.
- No copying, distribution, sharing, modification, reverse engineering, or derivative works allowed.
- No use for training other models, distillation, or knowledge extraction.
- No commercial use, sublicensing, or public release without explicit written permission from Within Us AI.
Any unauthorized use, reproduction, or distribution constitutes copyright infringement.
Full License
See the LICENSE file (recommended to upload) or contact Within Us AI for the complete legal terms.
By accessing or using this model, you agree to these terms.
- Downloads last month
- -
Safetensors
Model size
5B params
Tensor type
F32
·