SEN-AGI/Lura-1.1-500m

Lura 1.1: An Ultra-Lightweight, Efficient Language Model

Lura 1.1, developed by SEN-AGI, is a 0.5 billion parameter language model fine-tuned from the Qwen 2.5 500M base. It represents a significant advancement over its predecessor, Lura 1.0, by leveraging a unique data strategy. Instead of vast generic datasets, Lura 1.1 was trained on 4,000–5,000 custom synthetic samples, meticulously generated and filtered by frontier LLMs to maximize data density, diversity, and reasoning quality. This approach allows the compact model to achieve performance levels typically seen in larger models.

Key Capabilities & Improvements

• Enhanced Instruction Following: Demonstrates more reliable adherence to formatting, constraints, and multi-turn instructions.
• Improved Elementary Math & Logic: Better handles simple arithmetic and logical puzzles with step-by-step processing.
• Basic Code Generation: Capable of producing small scripts, fundamental functions, and code explanations.
• High Efficiency: Designed for smooth operation on resource-constrained environments such as CPUs, edge devices, smartphones, and systems with low VRAM.

Training & Development Highlights

Lura 1.1's training was notable for its "Zero-Cost Engineering" approach, conducted entirely on Google Colab's free T4 GPU tier and managed from a single mobile device. This highlights its potential for accessible and efficient model development.

Recommended Use Cases

Lura 1.1 is ideal for everyday basic assistance, simple text formatting and summarization, lightweight offline experimentation, and deployment on edge devices where computational resources are limited. Due to its compact architecture, users should be aware of its natural capacity limits, including potential for hallucinations on complex factual queries and limitations in advanced mathematics or deep multi-hop logic.