Microsoft Several new “Open” have launched the AI model Wednesday, the most capable of which is competitive with at least one criteria and 3-minute competition.
All-PHI 4 min arguments of new Pimisivoli licensed models, PHI 4 logic, and PHI 4 rational plus- “logic” model, which means they are capable of spending more time to solve a more time for complex problems. They expanded the PHI “Small Model” family of Microsoft, which the company launched a year ago to provide a foundation to create apps on the edge of AI developers.
PHI 4 Ministry was trained on nearly 1 million synthetic math problems produced by the Chinese AI Startup DEPSEC R1 argument model. About 3.8 billion parameters in size, PHI 4 min rational applications are designed for educational applications, Microsoft says, like “embedded tutoring” on the lightweight device.
The parameters are fairly matched by a model problem solving skill and more parameter models usually perform better than low parameters.
The PHI 4 argument, a 14-billion-parameter model, was trained using the “high quality” web data from the aforementioned “high quality” web data as well as “revised demonstrations”. According to Microsoft it is best for math, science and coding applications.
As a PHI 4 reasonable plus, it is adapted as an argument model for achieving better accuracy in specific tasks before Microsoft. Microsoft has claimed that PHI 4 argument has reached the performance level of Plus R1, a model with more parameters (671 billion) significantly. Fee 4 reasonable plus matching and 3-minute matching Omnith, math skills tests in the company’s internal benchmarking.
PHI 4 min argue, PHI 4 logic and PHI 4 argument plus available AI Dev platform hugged face With detailed technical reports.
TechCrunch event
Berkeley, CA
|
June 5
Book now
“They are used to“ learning to reinforce, and use high-quality data [new] Models are the size and the balance of performance, “Microsoft wrote Blog postThe “They are small enough for the short-latensi environment but maintains strong argument capabilities that compete with many large models this mixture also allows this mixture to perform complex rational tasks efficiently.”