Heaptalk, Jakarta — Microsoft unleashed a small AI model called Phi-3, offering users cost-effective options to build generative AI applications (04/23). Phi-3 comprises a set of small language models (SLMs), with Phi-3-mini as the first release.
Phi-3-mini contains 3.8 billion parameters and is available on Microsoft Azure AI Studio, Hugging Face, and Ollama. Microsoft developed this tiny AI model in two context-length variants: 4K and 128K tokens. Phi-3-mini is instruction-tuned, meaning that the model has been trained to follow instructions that reflect how people usually communicate.
The company claimed that Phi-3 models outperform models of the same size and next size up across various benchmarks that evaluate language, coding, and math capabilities. “Some customers may only need small models, some will need big models, and many are going to want to combine both in a variety of ways,” said Vice President of AI at Microsoft Luis Vargas (04/23).
The selection of an appropriate language model depends on an organization’s specific needs, task complexity, and available resources. Small language models are well suited for organizations looking to build applications that can run locally on a device instead of the cloud, where a task does not require extensive reasoning or a quick response. Conversely, large language models are more suited for applications that require the orchestration of complex tasks involving advanced reasoning, data analysis, and understanding of context.
Other SLMs in the Phi-3 family, including Phi-3 small with 7 billion parameters and Phi-3-medium with 14 billion parameters, will be launched in the coming weeks.
