🧑‍🏫 DeepNarra Hiraya-1-Instruct 1.0 (0.91B-A100M Super-Agent)

Phase 3 Instruction & Agentic Alignment — 14-Horizon Streaming Curriculum, CUA/MUA Action Tokens & DoReMi Balancing

Organization: DeepNarra
Pipeline Stage: Phase 3 — 14-Horizon Supervised Fine-Tuning (SFT) & Agent Alignment
Architecture: H-MoE-Zamba2-92-Expert-Nested-Hierarchical (908,645,372 total parameters, ~100M active/token)
Base Model Fine-Tuned: DeepNarra/Hiraya-1-H-MoE
Evaluation Governance: Zero Fake Scores Policy — 100% Verifiable Engineering Facts Only

---

🏛️ Verified Technical Specifications (Exact Repository Match)

Architectural Metric	Exact Verified Value in This Repository
Official Model Name	DeepNarra Hiraya-1-Instruct 1.0 (0.91B-A100M)
Exact Safetensors Parameter Count	908,645,372 parameters (3,323.94 MB / 3.32 GB in model.safetensors, 1,331 tensors)
Active Parameters / Token	~100M Active per Token (92 Sub-Experts per layer + Persistent Shared Foundation Expert)
Streaming SFT Execution	250 packed iterations across 227 unique cryptographic fingerprints (0% padding waste, completed in 49.6s)
Optimization Engine	Sophia 2nd-Order Clipped Hessian Optimizer + OHEM Adaptive Loss Gate + Focal Loss Scaling
14-Horizon Capabilities	English Fluency, Multi-Language Polyglot, Algorithms/Code, Math, JSON Tool Calling, CUA (<click>), MUA (<touch>), Security Auditing

---

🌐 The 14-Horizon Alignment Registry (config.json Verified)

1. English Fluency (FineWeb-Edu & Tulu-3)
2. All Code & Algorithms (StarCoder2 / Stack v2 & SmolLM Python-Edu)
3. Deep Multi-Step Reasoning (Open-R1 & GSM8K)
4. Agentic Tool Calling & JSON (No-Robots & Tulu-3)
5. Visual & Document Perception (LLaVA-Instruct)
6. Strict Constraints & Discipline (No-Robots / IFEval)
7. Long-Context Working Memory (Tulu-3 & FineWeb)
8. Hard Sciences & Medicine (SciQ & PubMedQA)
9. Global Multilingual Polyglot (Aya / Tulu-3 Polyglot)
10. Cybersecurity & Secure Coding (CodeFeedback Vulnerability Auditing)
11. Finance, Law & Contracts (FinQA & LegalBench)
12. Creative Writing & Nuance (UltraChat-200k)
13. Computer-Using Agent (CUA) (ScreenSpot / Desktop <click>坐标)
14. Mobile-Using Agent (MUA) (Mobile Actions <touch> UI Navigation)

---

🚀 Quickstart Usage (Python)

from transformers import AutoTokenizer
from safetensors.torch import load_file
from huggingface_hub import hf_hub_download

repo_id = "DeepNarra/Hiraya-1-Instruct"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
weights_path = hf_hub_download(repo_id=repo_id, filename="model.safetensors")
state_dict = load_file(weights_path)
print(f"Loaded {len(state_dict):,} Instruct H-MoE 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.