Skip to main content

SD Times Open-Source Project of the Week: Phi-3

Phi-3 is a family of open source small language models developed and made available by Microsoft. 

“Small language models are designed to perform well for simpler tasks, are more accessible and easier to use for organizations with limited resources, and they can be more easily fine-tuned to meet specific needs. They are well suited for applications that need to run locally on a device, where a task doesn’t require extensive reasoning and a quick response is needed,” Misha Bilenko, corporate vice president for Microsoft GenAI, wrote in a blog post. 

The idea behind developing a model so small was inspired by Microsoft researcher Ronan Elden reading a bedtime story to his daughter, which led him to think “how did she learn this word? How does she know how to connect these words?”

Applying this to AI, Elden wondered what would happen if an AI model was trained just on words that would be understood by a 4-year-old. 

Phi-3 comes in a variety of options: 

  • Phi-3-vision is a 4.2B parameter model that capable of understanding both text and vision
  • Phi-3-mini is a 3.8B parameter model, available in 128K and 4K context length options
  • Phi-3-small is a 7B parameter model, available in 128K and 4K context length options
  • Phi-3-medium is a 14B parameter model, available in 128K and 4K context length options

Phi-3-vision is the first multimodal model in the family, and can generate insights from charts and diagrams. “Phi-3-vision builds on the language capabilities of the Phi-3-mini, continuing to pack strong language and image reasoning quality in a small model,” Bilenko wrote. 

According to Microsoft, compared to other models, Phi-3 performs well. For example, Phi-3-small beats GPT-3.5T across a variety of language, reasoning, coding, and math benchmarks, while Phi-3-medium beats out Gemini 1.0 Pro. Additionally, Phi-3-vision outperforms Claude-3 Haiku and Gemini 1.0 Pro V in general visual reasoning tasks, OCR, table, and chart understanding tasks. 

All of the Phi-3 models are currently available on Azure AI and Hugging Face. 

The post SD Times Open-Source Project of the Week: Phi-3 appeared first on SD Times.



from SD Times https://ift.tt/AquJ7KN
Labels:

Comments

Post a Comment

Popular posts from this blog

Difference between Web Designer and Web Developer Neeraj Mishra The Crazy Programmer

Have you ever wondered about the distinctions between web developers’ and web designers’ duties and obligations? You’re not alone! Many people have trouble distinguishing between these two. Although they collaborate to publish new websites on the internet, web developers and web designers play very different roles. To put these job possibilities into perspective, consider the construction of a house. To create a vision for the house, including the visual components, the space planning and layout, the materials, and the overall appearance and sense of the space, you need an architect. That said, to translate an idea into a building, you need construction professionals to take those architectural drawings and put them into practice. Image Source In a similar vein, web development and design work together to create websites. Let’s examine the major responsibilities and distinctions between web developers and web designers. Let’s get going, shall we? What Does a Web Designer Do?
Post a Comment
Read more

A guide to data integration tools

CData Software is a leader in data access and connectivity solutions. It specializes in the development of data drivers and data access technologies for real-time access to online or on-premise applications, databases and web APIs. The company is focused on bringing data connectivity capabilities natively into tools organizations already use. It also features ETL/ELT solutions, enterprise connectors, and data visualization. Matillion ’s data transformation software empowers customers to extract data from a wide number of sources, load it into their chosen cloud data warehouse (CDW) and transform that data from its siloed source state, into analytics-ready insights – prepared for advanced analytics, machine learning, and artificial intelligence use cases. Only Matillion is purpose-built for Snowflake, Amazon Redshift, Google BigQuery, and Microsoft Azure, enabling businesses to achieve new levels of simplicity, speed, scale, and savings. Trusted by companies of all sizes to meet
Post a Comment
Read more

2022: The year of hybrid work

Remote work was once considered a luxury to many, but in 2020, it became a necessity for a large portion of the workforce, as the scary and unknown COVID-19 virus sickened and even took the lives of so many people around the world.  Some workers were able to thrive in a remote setting, while others felt isolated and struggled to keep up a balance between their work and home lives. Last year saw the availability of life-saving vaccines, so companies were able to start having the conversation about what to do next. Should they keep everyone remote? Should they go back to working in the office full time? Or should they do something in between? Enter hybrid work, which offers a mix of the two. A Fall 2021 study conducted by Google revealed that over 75% of survey respondents expect hybrid work to become a standard practice within their organization within the next three years.  Thus, two years after the world abruptly shifted to widespread adoption of remote work, we are declaring 20
Post a Comment
Read more