In response to increasing user demand, many companies have developed open-source large language models (LLMs) Traditionally, LLMs have been complex and require powerful computational resources and servers. However, the rising need for more accessible solutions has led to the creation of compact LLMs that can fit even into your phone.
Many LLMs are also paid and rely on premium pricing models. Several companies have introduced open-source LLMs that can be optimized for mobile use to meet user demand for more accessible solutions. Here are five notable options:
5 Mobile-Friendly Open-Source LLMs
1. Phi-2 LLMs
Phi-2 can be quantized to lower bit-widths like 4-bit or 3-bit precision, significantly reducing the model size to around 1.17–1.48 GB. This makes it efficient for running on mobile devices with limited memory and computational resources. Trained on a large corpus of web data, Phi-2 excels in tasks involving common sense reasoning, language understanding, and logical reasoning.
2. Gemma 2B
Despite its small size, Gemma 2B delivers high performance. It uses a multi-query attention mechanism, which reduces memory bandwidth requirements during inference. This is particularly advantageous in on-device scenarios with limited memory bandwidth. Gemma 2B has a strong track record in language understanding, reasoning, and safety.
3. LLMaMA-2-7B
LLMaMA-2-7B can run on devices with 6GB+ RAM, making it a viable option for developers looking to create intelligent language-based features for smartphones. While it requires sufficient RAM and may not match the speed of cloud-based models, it offers a good balance for on-device applications.
4. Falcon-RW-1B
Falcon-RW-1B is designed for resource-constrained devices like smartphones. It can add conversational capabilities to the Falcon-RW-1B-Instruct-OpenOrca model, enhancing user engagement and expanding use cases. This model is particularly useful for providing accessibility in resource-constrained environments, such as smartphones.
5. StableLM-3B
StableLM-3B can be quantized to lower bit-widths, like 4-bit precision, reducing the model size to about 3.6 GB. This optimization allows it to run efficiently on smartphones. Interestingly, StableLM-3B has reportedly outperformed Stable’s own 7B StableLM-Base-Alpha-v2.
These open-source LLMs are paving the way for more accessible and efficient mobile applications, meeting the growing demand for powerful yet compact language models.
See also: Generative AI Revolutionizes The Oil And Gas Industry
