Welcome to the Falcon 3 Family of Open Models!

The Falcon 3 family of open models has been released, featuring three variants: Falcon 3-7B, Falcon 3-40B, and Falcon 3-80B, with model sizes of 7 billion, 40 billion, and 80 billion parameters, respectively. These models utilize a transformer architecture with advancements in tokenization and training techniques, achieving state-of-the-art performance on benchmarks such as MMLU and HELM. This release provides practitioners with scalable options for various applications in natural language processing, promoting accessibility and further research in the open-source community.