SmolLM2 Released: The New Series (0.1B, 0.3B, and 1.7B) of Small Language Models for On-Device Applications and Outperforms Meta Llama 3.2 1B

**Transforming Natural Language Processing with SmolLM2** Recent advancements in large language models (LLMs) like GPT-4 and Meta’s LLaMA have improved how we work with language tasks. However, these large models require a lot of computing power and memory, making them hard to use on devices like smartphones. Running them locally can also be expensive. This has created a need for smaller, efficient models that perform well on devices. **Introducing SmolLM2** Hugging Face has launched SmolLM2, a series of compact models designed for use on devices. Building on the success of SmolLM1, SmolLM2 offers better performance while being lightweight. It comes in three sizes: 0.1B, 0.3B, and 1.7B parameters. The main advantage is that these models can run directly on devices, removing the need for large cloud systems. This is perfect for situations where speed, privacy, and hardware limitations are important. **Compact and Versatile** SmolLM2 models are trained on a vast amount of data, focusing mainly on English text. They excel at tasks like text rewriting, summarization, and function calling, making them useful in areas with limited internet access. Performance tests show that SmolLM2 outperforms Meta Llama 3.2 1B and, in some cases, beats benchmarks set by Qwen2.5 1B. **Advanced Training Techniques** SmolLM2 uses advanced training methods, such as Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO). These techniques help the models follow complex instructions and provide accurate answers. They also work well with frameworks like llama.cpp and Transformers.js, allowing for efficient use on local CPUs or browsers without needing special GPUs. This flexibility makes SmolLM2 great for edge AI applications, focusing on low latency and data privacy. **Significant Improvements Over SmolLM1** SmolLM2 represents a step forward in making powerful LLMs more accessible for various devices. Compared to SmolLM1, which had limitations, SmolLM2 shows significant improvements, especially in the 1.7B version. It supports advanced features like function calling, making it useful for automated coding and personal AI applications. **Impressive Benchmark Results** Benchmark scores show that SmolLM2 has enhanced performance, often matching or exceeding that of Meta Llama 3.2 1B. Its compact design allows it to work effectively where larger models struggle, making it essential for industries concerned about costs and the need for real-time processing. **Efficient and Versatile Solutions** SmolLM2 is built for high performance, with sizes ranging from 135 million to 1.7 billion parameters, balancing versatility and efficiency. It handles text rewriting, summarization, and complex functions while improving mathematical reasoning—making it a cost-effective choice for on-device AI. As small language models become more popular for privacy-focused and latency-sensitive applications, SmolLM2 sets a new standard in on-device natural language processing. **Explore SmolLM2 and Let AI Transform Your Business** Discover how SmolLM2 can enhance your operations. Identify automation opportunities, set measurable goals for your AI projects, choose the right solutions, and implement them step by step. For guidance on managing AI KPIs, contact us at hello@itinai.com. For insights on leveraging AI, follow us on Telegram or Twitter. Experience how AI can improve your sales processes and increase customer engagement. Explore our solutions at itinai.com.