Microsoft launches lightweight model Phi-3 mini that can run on smartphones

The model can also run on the chips that power regular computer instead of the expensive and hard to find AI processors that Nvidia makes.

April 24, 2024 02:46 pm | Updated 02:47 pm IST

FILE PHOTO: Microsoft launched Phi-3 Mini, a freely available, lightweight small language AI model.

FILE PHOTO: Microsoft launched Phi-3 Mini, a freely available, lightweight small language AI model. | Photo Credit: Reuters

Microsoft on Tuesday launched Phi-3 Mini, a freely available, lightweight small language AI model. The company plans to further release two other models in the same scale. The model is meant to be simpler and cheaper compared to other heavyweight large language models like OpenAI’s GPT-4 Turbo. This is the smallest model that Microsoft has released yet and can be run locally, which means it can run for free on a smartphone without requiring an internet connection. 

The model can also run on the chips that power regular computer instead of the expensive and hard to find AI processors that Nvidia makes.  

Phi-3 mini has 3.8 billion parameters while the two other models that are yet to be released -- Phi-3 Small and Phi-3 Medium -- will have 7 billion parameters and 14 billion parameters respectively. The Phi-3 mini will be available on Azure, Hugging Face and Ollama. 

The company had released Phi-2 in December that performed on par with bigger models like Llama 2. Microsoft claims that Phi-3 performs better than this older version and can respond to queries in a manner that is close to a model 10 times bigger its size. 

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

Eric Boyd, corporate vice president at Microsoft Azure AI Platform, told The Verge that Phi-3 Mini is as capable as ChatGPT, but “in a smaller form factor.”

 The Information had reported early this year that Microsoft was building a team that was focused on making lighter-weight AI models. Aside from the Phi range of models, the company has also built a model called Orca-Math that just solves math problems. 

The ongoing large language model race has moved from massive models to smaller ones that perform specific tasks given the compute and costs that bigger AI models need. Specialised models can also be trained to be more accurate than the large language models that are trained to perform all kinds of tasks. 

Google’s Gemma models excel at being chatbots and language-based tasks while Meta’ Llama 3-8 billion model is good at coding assistance and Anthropic’s Claude 3 Haiku can summarise lengthy documents quickly. Smaller startups like AI21 Labs have also recently launched an AI-powered writing assistant based on a specialised AI model called Jurassic-2.

 Boyd told the tech outlet that while Phi-1 was just focused on coding and Phi-2 on reasoning, Phi-3 was built to be better at both. 

0 / 0
Sign in to unlock member-only benefits!
  • Access 10 free stories every month
  • Save stories to read later
  • Access to comment on every story
  • Sign-up/manage your newsletter subscriptions with a single click
  • Get notified by email for early access to discounts & offers on our products
Sign in

Comments

Comments have to be in English, and in full sentences. They cannot be abusive or personal. Please abide by our community guidelines for posting your comments.

We have migrated to a new commenting platform. If you are already a registered user of The Hindu and logged in, you may continue to engage with our articles. If you do not have an account please register and login to post comments. Users can access their older comments by logging into their accounts on Vuukle.