Falcon open models land on Hugging Face: Falcon-7B/40B and Instruct variants under Apache 2.0

AI Impact Summary

Hugging Face now hosts the Falcon family from TI, including Falcon-7B, Falcon-40B, and their Instruct variants under Apache 2.0. The models emphasize open access with quantization and 8-bit inference options, and rely on remote code execution for some setups, which affects deployment tooling and security review. The availability of open-weight models and integration with Text Generation Inference and Core ML weights enables rapid experimentation on HF workflows, but production-grade use will require substantial GPU memory (Falcon-40B ~90GB; 7B ~15GB) and careful validation of licensing and inference paths.