Our Claudia Sacco, Professional Solutions Architect, and Andrea Policarpi, Data Scientist at BIP xTech, co-authored an insightful article published on the AWS Blog. It outlines the technical journey Fastweb undertook to build an Italian-specific large language model.
AI Innovation: the fine-tuning of the mistral model
The article discusses how Fastweb, one of Italy’s leading telecommunications providers, leveraged Amazon SageMaker HyperPod to fine-tune the Mistral model. This was a key step in their effort to create a large language model (LLM) trained on Italian data. By adapting existing AI models to the specific needs of the local market, Fastweb is positioning itself at the cutting edge of AI technology in Italy.
Fine-tuning a language model is no small task. It requires high-performance computing power, vast data processing capabilities, and cutting-edge AI services. Amazon SageMaker HyperPod enabled Fastweb to provision resilient clusters at scale, leveraging the power of AWS Trainium and NVIDIA A100 GPUs. This flexible infrastructure allowed Fastweb to maintain an agile and cost-effective process, crucial for managing complex AI workloads.
A transformative step for the italian AI landscape
Fastweb’s initiative represents a significant milestone in Italy’s AI ecosystem. The goal is to create high-performing Italian-language LLMs for broader use. This project demonstrates how machine learning models can be customized for regional demands. It also opens up new opportunities for AI-driven innovation in Italy.
