Hugging Face, an AI platform company, has released SmolLM2, a family of compact language models available under the Apache 2.0 license. The models come in three sizes: 135 million, 360 million, and 1.7 billion parameters, designed to run on smartphones and Edge devices.
SmolLM2 supports text rewriting, summarization, and function-calling capabilities. The largest variant, trained on 11 trillion tokens, uses diverse datasets, including FineWeb-Edu and specialized mathematics and coding datasets. It outperforms Meta's Llama 1 billion model on several benchmarks, scoring 6.13 on MT-Bench for chat capabilities and 48.2 on the GSM8K benchmark for mathematical reasoning.
Analyst QuickTake: This follows the company’s previous launch of the SmolLM model in July of this year. This new iteration enhances the capabilities of on-device AI, allowing users to run powerful models directly on their smartphones. The SMOLLM2 model claims improved efficiency and performance, making it easier for developers to integrate sophisticated AI features into mobile applications.
By using this site, you agree to allow SPEEDA Edge and our partners to use cookies for analytics and personalization. Visit our privacy policy for more information about our data collection practices.