YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
library_name: transformers tags:
qwen3
slm
cognitive-ai
logic
verbarex license: apache-2.0 base_model: Qwen/Qwen3-1.7B language:
en
LuminoLex-1.7B (Quantum 4D Edition)
LuminoLex-1.7B is a high-performance Small Language Model (SLM) engineered by VERBAREX. It leverages a unique 4D Cognitive Architecture to deliver reasoning capabilities typically found in much larger models.
Model Details
Model Description
LuminoLex-1.7B is built upon the Qwen-3 architecture but undergoes a deep adaptation process using high-rank LoRA and the NEFTune technique. It is designed to be a "Pure Brain" model, operating in native Float16 to maintain maximum logical fidelity.
Developed by: VERBAREX
Model type: Causal Language Model (SLM)
Language(s) (NLP): English
License: Apache 2.0
Finetuned from model: Qwen/Qwen3-1.7B
Model Sources
Repository: VERBAREX/LuminoLex-1.7B
Architecture: 4D Cognitive Architecture (ACL, QPB, HIM, TLF)
Uses
Direct Use
LuminoLex is optimized for complex reasoning, mathematical problem solving, and algorithmic code generation. It is intended for deployment in environments where low latency and high cognitive density are required.
Out-of-Scope Use
This model is not intended for high-stakes medical or legal advice without human oversight. Despite its advanced reasoning, it remains a 1.7B parameter model and may exhibit limitations in broad world-knowledge retrieval compared to LLMs.
Bias, Risks, and Limitations
LuminoLex is trained to be factually grounded; however, users should be aware of potential hallucinations in niche data areas. The "Quantum Probability Branching" attempts to mitigate logic errors, but verification is recommended for mission-critical code.
How to Get Started with the Model
from transformers import AutoModelForCausalLM, AutoTokenizer import torch
model = AutoModelForCausalLM.from_pretrained( "VERBAREX/LuminoLex-1.7B", torch_dtype=torch.float16, device_map="auto" )
Training Details
Training Data
The model utilizes a Tri-Core Balanced Dataset strategy:
General Logic: UltraChat (Conversational flow)
Mathematics: Orca-Math (Step-by-step reasoning)
Coding: Evol-Instruct-Code (Programming logic)
Training Procedure
Training Hyperparameters
Precision: Full Float16 (non-mixed)
Optimizer: AdamW
Learning Rate: 2e-4 (Cosine schedule)
LoRA Rank: 64 (Alpha: 128)
NEFTune Noise Alpha: 5
Technical Specifications
Model Architecture and Objective
LuminoLex integrates an experimental 4D Cognitive Architecture:
Autonoetic Consciousness Layer (ACL): Self-verification of identity and constraints.
Quantum Probability Branching (QPB): Parallel path evaluation for logic.
Holographic Intent Mesh (HIM): Nuanced intent analysis.
Temporal Logic Folding (TLF): Future-state projection for safer logic.
Compute Infrastructure
Software
Transformers, PEFT, PyTorch.
More Information
Developed by VERBAREX to push the boundaries of Small Language Models through cognitive layer injection.
Model Card Contact
For inquiries, contact the VERBAREX research team.