Model Overview
The somosnlp/GemmaColRAC-AeroNavigatorLLM-2b model, developed by Edison Bejarano, Sergio Nicolas, and Santiago Pineda, is a fine-tuned version of google/gemma-2b-it. This 2.5 billion parameter model is specifically engineered to improve the comprehension and accessibility of the Colombian Aeronautical Regulations (RAC) in Spanish.
Key Capabilities
- Aeronautical Regulation Interpretation: Designed to simplify the understanding and application of complex aviation regulations.
- Text Generation: Capable of generating text to provide comprehensible information extracted from the RAC.
- Specialized Domain Knowledge: Fine-tuned on the
RAC_Colombia_QualityImproved025 dataset, a high-quality, improved version of the Colombian Aeronautical Regulations.
Training Details
The model was trained on a Tesla V100-SXM2-16GB GPU for approximately 70 minutes. It utilized a learning rate of 0.00005 with a Paged AdamW 8bit optimizer over 258 maximum steps. The training process focused on efficiency, resulting in an estimated CO2 emission of 0.0741 kg.
Use Cases
This model is ideal for direct applications requiring the interpretation and application of aeronautical regulations. It serves professionals and enthusiasts in the aviation sector by providing immediate and understandable access to information derived from the RAC, making complex regulatory texts more approachable.