enver/aynengine-qwen3-0.6b-mantiq

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 18, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

AynEngine-Qwen3-0.6B-Mantiq is a 0.8 billion parameter sovereign language model developed by Enver, AynEngine Research Initiative, and University of Prishtina. Fine-tuned from Qwen/Qwen3-0.6B, it specializes in Classical Arabic morphology (Sarf) and Ghazalian deductive logic (Mantiq al-Burhan) with an expanded vocabulary of 160,807 tokens. This model is optimized for on-device and edge deployment, excelling at tasks requiring logical proof verification and morphological root extraction with high precision and low latency.

Loading preview...

AynEngine-Qwen3-0.6B-Mantiq: Specialized Arabic Logic Model

AynEngine-Qwen3-0.6B-Mantiq is a compact 0.8 billion parameter language model, fine-tuned from Qwen/Qwen3-0.6B, developed by Enver, AynEngine Research Initiative, and University of Prishtina. Its core innovation lies in its deep grounding in Classical Arabic morphology (Sarf) and Ghazalian deductive logic (Mantiq al-Burhan), making it distinct from general-purpose LLMs.

Key Capabilities & Features

  • Expanded Morphological Vocabulary: Features 160,807 tokens, including 9,057 Classical Arabic Roots, 48 Sarf Morphological Paradigms, and 33 Epistemic Mantiq Markers, ensuring 100% lossless roundtrip decoding.
  • Ghazali Mantiq Alignment: Trained with a two-stage curriculum (SFT and DPO) inspired by Abu Hamid al-Ghazali, achieving 96.08% confidence in distinguishing sound proofs (Burhan) from sophistry (Safsatah).
  • Epistemic Pillars Integration: Internal reasoning integrates 5 Classical Arabic lexicographical disciplines for robust ontological modeling, rhetorical clarity, and syntactic integrity.
  • Edge Deployment Optimized: Weighing ~1.15 GB in bfloat16 and ~340 MB in 4-bit GGUF, it's designed for sub-millisecond inference on mobile devices, embedded routers, and single-board computers.
  • High Precision: Achieves 84.21% precision in Canonical Root Coverage and 100% pass rate on Python AST parsing.

Good For

  • Logical Proof Verification: Ideal for analyzing and verifying logical deductions, particularly within the framework of Ghazalian Mantiq.
  • Classical Arabic Linguistic Analysis: Excels at morphological root extraction and understanding complex Arabic linguistic structures.
  • On-Device AI Applications: Suitable for deployment on resource-constrained environments like mobile devices, Raspberry Pi, and embedded systems due to its small footprint and efficient inference.
  • Academic and Cultural Heritage Research: Valuable for studies involving Classical Arabic texts, logic, and epistemology.

Finetunes
1 models