🌲 DeepNarra Narra 1.0 (28.5M-A6.2M Sovereign Edge)

The H-MoE Bedrock Tree — Strictly Locked 128K Context (131,072) & Lyapunov GLA Recurrence

Organization: DeepNarra
Designation: The H-MoE Bedrock Tree (Deep Roots, Strong Trunk, Spreading Branches)
Architecture: DeepNarraNarraHMoEForCausalLM — Chunk-wise Gated Linear Attention (GLA) + Lyapunov Stability + 92 Sub-Experts
Inviolable Context Law: Strictly Locked to 128K (131,072 tokens)
Empirically Measured Velocity: 325.2 tok/s (49.20 ms latency) (Verified live on single Tesla T4 GPU)
Evaluation Governance: Zero Fake Scores Policy — 100% Verifiable Engineering Facts Only

---

       🌲 NARRA                          🦈 MAKO                          🍎 HIRAYA
 The Tree Anchored in the Sand     The Water Where the Shark Thrives    The Fruit of All That Work
(Unyielding Strength & Deep Roots)     (Fluid Speed & Freedom)       (Realization of Hopes & Dreams)
                                     ⚡
                           👑 DEEPNARRA PAHAM (THE SAGE)
                     (The Sovereign Final Frontier)

---

🏛️ Verified Technical Specifications (Exact Repository Match)

Architectural Metric	Exact Verified Value in This Repository
Official Model Name	DeepNarra Narra 1.0 (28.5M-A6.2M Sovereign Edge)
Exact Safetensors Parameter Count	28,519,602 parameters (108.94 MB in model.safetensors, 1,330 tensors)
Active Parameters / Token	~6.2M Active / Token (23 Macro-Engines × Top-2 Micro-Experts + Shared Foundation Expert)
Native Ternary GGUF Binary	5.85 MB (model-TQ1_0.gguf — BitNet b1.58 Ternary TQ1_0 at 1.58 bits/weight, 1,330 tensors)
Context Window Boundary	Strictly Locked to 131,072 tokens (128K) — Never altered or expanded
Measured Inference Throughput	325.2 tokens/sec (49.20 ms step latency on Tesla T4)
Recurrent Attention Backbone	Chunk-wise Gated Linear Attention (GLA) with O(1) constant inference memory
Lyapunov Stability Bound	Spectral norm & energy bounding (V(S_t) = Tr(S_t^T P S_t) ≤ 100.0)
Recurrent State Serialization	6.0 Bytes / Recurrent State via Adaptive Tapered Float Pool

---

⚡ Distinct Architectural Innovations (Unique to Narra)

1. Strictly Locked 128K Context Boundary (131,072 tokens):

   • Engineered specifically for rock-solid, drift-free long-document retention up to exactly 131,072 tokens without attention sink degradation.

2. Chunk-Wise Gated Linear Attention (GLA) with Lyapunov Bounding:

   • Updates recurrent memory state $S_t$ via hardware-aligned chunks: $$S_t = lpha_t \odot S_{t-1} + K_t^T V_t, \quad O_t = Q_t S_t$$
   • Enforces a strict Lyapunov stability ceiling so recurrent states never overflow or decay over 128K tokens.

3. 92-Branch H-MoE Canopy + Deep Root Shared Expert:

   • Mirrors the structure of the Philippine Narra tree: a single unshakeable Shared Foundation Expert trunk supporting 23 Macro-Branches and 92 specialized Micro-Leaves.

---

🚀 Quickstart Usage

1. Ollama (Native TQ1_0 GGUF + Included Modelfile)

ollama create deepnarra-narra -f Modelfile
ollama run deepnarra-narra

2. Python SafeTensors Inspection & Loading

from safetensors.torch import load_file
from huggingface_hub import hf_hub_download

weights_path = hf_hub_download(repo_id="DeepNarra/Narra", filename="model.safetensors")
state_dict = load_file(weights_path)
print(f"Loaded {len(state_dict)} tensors ({sum(v.numel() for v in state_dict.values()):,} parameters)")

---

⚖️ License, Dataset Attributions & Responsible AI Compliance

1. Model License (Apache-2.0)

This repository, including all model weights (.safetensors and .gguf), configuration manifests, and custom architectural code, is licensed under the Apache License, Version 2.0 (http://www.apache.org/licenses/LICENSE-2.0). See the LICENSE and NOTICE files in the root of this repository for full legal terms.

2. Upstream Dataset & Open-Data Attributions (ODC-By-1.0 & Apache-2.0)

In strict compliance with open-data attribution requirements:

• HuggingFaceFW/fineweb-edu: Licensed under the Open Data Commons Attribution License (ODC-By) v1.0.
• CohereForAI/aya_collection: Licensed under the Apache License 2.0.
• m-a-p/CodeFeedback-Filtered-Instruction: Licensed under the Apache License 2.0.
• HuggingFaceH4/no_robots: Attributed under CC-BY-NC-4.0 for instruction-following research evaluation.
• SymPy CAS Procedural Reasoning Stream: Synthetically generated and formally verified via SymPy (BSD-3-Clause).

3. Intended Use, Safety & Limitations

• Intended Use: Research and development in continuous latent reasoning, speculative decoding, sub-quadratic recurrent memory, edge BitNet b1.58 GGUF deployment, and sandboxed autonomous agents.
• Safety & Execution Airlock: Code execution and tool-calling capabilities are designed to run strictly inside the DeepNarra Airlock Sandbox with AST static verification and SSRF network firewalling (169.254.169.254 and RFC-1918 private subnets blocked).
• Limitations & Disclaimer: As an experimental sovereign architecture trained on compact compute budgets, outputs should be independently verified for high-stakes medical, legal, financial, or safety-critical systems. Provided "AS IS" without warranties of any kind under Section 7 of the Apache 2.0 License.