The embedded systems industry in Spain has hit significant strides in recent years on the back of pioneering advancements in artificial intelligence and supercomputing. One of the key players propelling this trend is Cerebras Systems, in collaboration with the Barcelona Supercomputing Center (BSC-CNS). Together, they have successfully completed the training of FLOR-6.3B, a cutting-edge English-Spanish-Catalan large language model (LLM). FLOR-6.3B represents a significant milestone in AI development, especially for languages with limited data resources like Catalan and Spanish. Despite Spanish being one of the most spoken languages worldwide, there is a notable scarcity of training data available online. Catalan, with even less digital presence, presents an additional challenge. To address this, BSC and Cerebras employed innovative AI training techniques, using a fully-trained LLM and adjusting the embedding layer to simulate the effect of training with a large dataset. The FLOR models, ranging from 760M to 6.3B parameters, are based on publicly released checkpoints of BLOOM. For the FLOR models, a new tokenizer was developed to better adapt to Catalan and Spanish. This tokenizer has a reduced vocabulary set, resulting in fewer parameters than the original Bloom-7.1B model, thereby reducing inference costs by over 10%. These advancements have significant implications for the embedded systems industry in Spain. The integration of sophisticated AI models like FLOR-6.3B into embedded systems can revolutionize various sectors, including telecommunications, biomedical applications, and industrial automation. The ability to efficiently process and analyze multilingual data enhances the functionality and versatility of embedded systems. In conclusion, the successful training and deployment of FLOR-6.3B increases the capabilities and competitiveness of Spain’s embedded systems industry on the global stage.