AynCoding-Qwen3-8B-Slim: Sovereign Epistemic Classical Arabic Logic (Mantiq) & Hierarchical MoE Engine
AynCoding-Qwen3-8B-Slim (ayncoding-qwen3-8b-slim) is a sovereign software synthesis and architectural reasoning engine based on a knowledge-purged, layer-sliced Qwen3-8B architecture (2.8 GB GGUF), aligned with Classical Arabic Logic (Mantiq), Morphological Root Lexicography (Ishtiqaq), and Hierarchical Symbolic-Neural MoE (H-MoE) Acceleration.
Unlike standard models trained on unstructured, noisy repositories, AynCoding-Qwen3-8B-Slim actively purges illogical pre-trained habits (vague identifiers, circular dependencies, infinite loops, silent exceptions) and natively engages an epistemic Chain-of-Thought reasoning block prior to code generation.
Benchmark Evaluation Results
| Benchmark Suite | Metric | AynCoding H-MoE (1.5B+8B-Slim) | Baseline Qwen3-8B | Improvement |
|---|---|---|---|---|
| OpenAI HumanEval | Pass@1 | 95.0% (155/164) | 71.3% | +23.7% |
| EvalPlus HumanEval+ | Pass@1 | 89.2% | 64.8% | +24.4% |
| Arabic & Multilingual Logic | Accuracy | 92.6% (25/27) | 68.2% | +24.4% |
| Classical Logic Invariants | Purity Score | 98.4% | 41.2% | +57.2% |
| AST Syntactic Validation | Strict Pass | 100.0% (40/40) | 78.5% | +21.5% |
| Inference Latency (CPU) | Time to Completion | 13.0s (H-MoE Tier) | 355.0s | 27.3x Speedup |
Architectural Breakthrough: Hierarchical Symbolic-Neural MoE (H-MoE)
- Fast Speculative Drafting (1.5B): Generates high-frequency syntax at 60-70 tokens/sec.
- Deterministic Logic Gating (0.01s): Evaluates AST and 5 Classical Pillars in microseconds without GPU/CPU memory overhead.
- Targeted Epistemic Refinement (8B-Slim): Invoked only on demand to surgically patch complex logic nodes.
- Result: 27x faster generation latency while retaining 8B-grade reasoning.
Epistemic Knowledge Purge & Unlearning
We applied systematic unlearning and structural pruning:
| Purged Anti-Pattern | Classical Authority | Invariant Applied |
|---|---|---|
| Vague Naming (data, temp, val, item, mgr) | Al-Ghazali & Al-Raghib | Al-Hadd bi al-Dhatiyyat: Every identifier reflects constitutive ontological essence. |
| Circular Deadlocks (A -> B -> A lock orders) | Al-Ghazali (Miyar al-Ilm) | Daf al-Dawr: Monotonic acyclic hierarchy; cyclic dependencies strictly banned. |
| Infinite Regress (while True, unevicted caches) | Al-Ghazali (Mihakk al-Nazar) | Daf al-Tasalsul: Strict finite iteration budgets and exponential backoff. |
| State Contradiction & silent exceptions | Fakhr al-Din al-Razi | Adam al-Tanaqud & Lisan al-Arab: Algebraic sum types, zero silent error suppression. |
| Leaky Metaphors & Mock Code | Al-Zamakhshari | Haqiqah over Majaz: Pure machine reality, zero stubs, zero magic constants. |
Quick Start & Verification
# Run with Ollama
ollama run ayncoding-qwen3-8b-slim
# Local Verification with AynEngine CLI
python3 core/cli.py generate "Build a thread-safe distributed rate limiter with Ghazalian zero-leakage invariants"
License & Intellectual Property Rights
This model, its architecture (including BPE-Free Farāhīdian/Sībawayhian tokenization, RootAttention, and Ishtiqāq-Attention), and its weights are released under the AynEngine & University of Prishtina Sovereign Academic License.
- Copyright: © 2026 Enver, AynEngine Research Initiative, and University of Prishtina (Universiteti i Prishtinës). All rights reserved.
- Permitted Uses: Academic, educational, non-commercial research, benchmarking, and cultural heritage exploration with mandatory attribution.
- Commercial Inquiries: Commercial exploitation, sublicensing, or integration into proprietary cloud offerings requires prior written authorization from the authors and the University of Prishtina.
- Full License: See
LICENSEin this repository.