| --- |
| language: |
| - en |
| license: apache-2.0 |
| pipeline_tag: text-generation |
| tags: |
| - recursive-ai |
| - opaque-reasoning |
| - reasoning-model |
| - sovereign-ai |
| - frontier-ai |
| - recursive-cognition |
| - agentic-ai |
| - hybrid-mind |
| - multimodal |
| - long-context |
| - rlhf-ready |
| - safetensors |
| - withinusai |
| - Royal.Opaque.Reasoner.IX |
| model_type: ror_ix |
| --- |
| |
| 👑 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.” |
