Hugging Face and Meta Release Compact AI Models for Mobile Devices
SmolLM2 and MobileLLM aim to bring efficient language processing to smartphones and edge devices, offering high performance with reduced resource demands.
- Hugging Face's SmolLM2 series features models ranging from 0.1B to 1.7B parameters, optimized for on-device applications with improved instruction handling and privacy.
- SmolLM2 models outperform Meta's Llama 3.2 1B in benchmarks and offer advanced functionalities like function calling, enhancing their utility for personal AI applications.
- Meta's MobileLLM, with models up to 1B parameters, focuses on depth over width to maximize performance on mobile devices, emphasizing efficient architecture over size.
- MobileLLM employs techniques like embedding sharing and grouped query attention to optimize performance while minimizing latency and energy consumption.
- Both model series are available on Hugging Face, with SmolLM2 under an Apache 2.0 license and MobileLLM under a Creative Commons 4.0 non-commercial license, promoting research and development.
MarkTechPost 
VentureBeat 
