Microsofttoday announced Phi-4, a 14B-parameter "state-of-the-art" small language model (SLM) that, in addition to traditional language processing, excels atmath and other fieldscomplex inferencePhi-4 is the latest addition to the Phi family of small language models, and officials say it demonstrates Microsoft's continued exploration of the possibilities of SLM boundaries.
Officials say it benefits from technological advances in a variety of areas, including the use ofHigh-quality synthetic datasets, carefully selected high-quality organic data, as well as post-training innovations, Phi-4 outperforms mathematical reasoning in theModels of the same type and on a larger scaleThe results of the benchmarks are summarized in the following table. Its performance on the math competition problem outperformed several larger scale models including Gemini Pro 1.5.1AIttached is a technical paper on the benchmarking results:Click here to go
Microsoft has announced that it is making "powerful and responsible" AI capabilities available to all customers using the Phi family of models, including Phi-3.5-mini. Phi-4 is now available on Azure AI Foundry.