🍎 DeepNarra Hiraya 1.0 (28.5M-A6.2M Sovereign Edge)

The Sovereign Fruit & Flagship ("Thinker & Faster") — Continuous Latent Deliberation & Symplectic H-MoE

Organization License Params GGUF

Organization: DeepNarra
Designation: The Sovereign Fruit & Flagship ("Thinker & Faster")
Architecture: DeepNarraHirayaForCausalLM — Two-Tier 92-Expert H-MoE + 768-D Coconut Latent Space + Symplectic Störmer-Verlet
Full-Scale 0.91B H-MoE Checkpoint: DeepNarra/Hiraya-1-Reasoning (908,645,372 params)
Empirically Measured Velocity: 378.5 tok/s (42.27 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 Hiraya 1.0 (28.5M-A6.2M Sovereign Edge)
Exact Safetensors Parameter Count 28,520,370 parameters (108.95 MB in model.safetensors, 1,331 tensors)
Active Parameters / Token ~6.2M Active / Token (Edge Scale) | ~100M Active / Token (0.91B Full Scale)
Native Ternary GGUF Binary 5.85 MB (model-TQ1_0.gguf — BitNet b1.58 Ternary TQ1_0 at 1.58 bits/weight, 1,331 tensors)
Measured Inference Throughput 378.5 tokens/sec (42.27 ms step latency on Tesla T4)
Continuous Latent Phase Space 768-D Coconut Continuous Latent Reasoning with explicit <think> and <latent_thought> scratchpads
Hamiltonian Latent Physics Symplectic Störmer-Verlet Integrator ($\Delta E < 10^{-13}$ energy conservation bound)
Geometric Manifold Projection Minkowski Lorentz Hyperboloid Embeddings + Langevin Stochastic Exploration
Hierarchical MoE Structure 23 Macro-Engines × 4 Micro-Experts = 92 Sub-Experts per layer + Persistent Shared Foundation Expert

⚡ Distinct Architectural Innovations (Unique to Hiraya)

  1. 768-Dimensional Coconut Continuous Latent Deliberation:

    • Evolves continuous thought vectors inside ℝ⁷⁶⁸ (768-D) prior to discrete token emission, preventing premature verbal commitment on complex math and coding tasks.
  2. Symplectic Störmer-Verlet Dynamics & Minkowski Hyperboloid Geometry:

    • Preserves Hamiltonian phase-space volume across multi-hop reasoning steps: $$p_{t+1/2} = p_t - rac{\eta}{2} abla_q U(q_t), \quad q_{t+1} = q_t + \eta p_{t+1/2}, \quad p_{t+1} = p_{t+1/2} - rac{\eta}{2} abla_q U(q_{t+1})$$
    • Projects hierarchical concepts onto the Lorentz hyperboloid (-x_0² + ∑ x_i² = -1/K) for distortion-free tree embeddings.
  3. Two-Tier 92-Expert H-MoE with Persistent Shared Foundation Expert:

    • Every token activates the Persistent Shared Foundation Expert plus Top-K routed domain experts across 23 specialized Macro-Engines (1,331 total weight tensors).

🚀 Quickstart Usage

1. Ollama (Native TQ1_0 GGUF + Included Modelfile)

ollama create deepnarra-hiraya -f Modelfile
ollama run deepnarra-hiraya

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/Hiraya", 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.
Downloads last month
374
Safetensors
Model size
28.5M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for DeepNarra/Hiraya

Quantized
(1)
this model

Datasets used to train DeepNarra/Hiraya