Modelpublic

waveforce-ai/Aura-1-Coding

sourceHugging Facellama3.1updated 11d agoView on Hugging Face

0likes

Model Card

Aura-1-Coding (Standalone 4-Bit Variant)

Aura-1-Coding is an ultra-fast, standalone code generation model optimized using localized high-density distillation loops. Built on top of the Meta-Llama-3.1-8B-Instruct architecture, it features a protected, internal chain-of-thought system designed to execute complex operations without raw reasoning exposure, preventing multi-level distillation attacks and consumer dilution.

⚡ Performance Footprint & Specifications

The primary execution architecture operates under the following hardware parameters:

Metric / Component	Configuration Specification
Base Architecture	Meta-Llama-3.1-8B-Instruct
Quantization Format	4-Bit NormalFloat (`NF4`) with Double Quantization
Compute Data Type	`Float16` Execution Gates
Optimizer Blueprint	8-Bit Paged Memory Managed
Target Alignment	High-Density Multi-File Coding Stack
VRAM Operational Footprint	~5.5 GiB (Ideal for consumer-grade GPU pipelines)

📁 Repository Structure

The primary model assets are isolated inside the main distribution branch:

text

waveforce-ai/Aura-1-Coding/
└── aura-1-coding-model-file/
    └── model.safetensors  <- [5.70 GB Standalone Weights Binary]