Microsoft Phi-3-vision Benchmark: Comparable to Claude 3-haiku and Gemini 1.0 Pro
On May 28, Microsoft unveiled the latest member of the Phi-3 family, the Phi-3-vision, at the Build 2024 conference. This model focuses on “visual capabilities,” understanding both text and images and is designed for efficient operation on mobile platforms.
The Phi-3-vision is a multimodal small language model (SLM) with 4.2 billion parameters and a context length of 128k tokens, supporting regular visual reasoning tasks and other functions. Microsoft published a new paper today stating that this SLM matches models like Claude 3-haiku and Gemini 1.0 Pro.
In their paper, Microsoft compared Phi-3-vision with models like ScienceQA, MathVista, and ChartQA. Despite having fewer parameters, Phi-3-vision performs exceptionally well.
Previous reports highlighted comparisons between Phi-3-vision and other models such as ByteDance’s Llama3-Llava-Next (8B), the Microsoft Research and University of Wisconsin, Columbia University collaboration LlaVA-1.6 (7B), and Alibaba’s QWEN-VL-Chat model, showcasing Phi-3-vision’s strong performance in various tasks.
Microsoft has uploaded this model to Hugging Face, inviting interested users to visit the project page for more information.
Keep visiting for more such awesome posts, internet tips, lifestyle tips, and remember we cover,
“Everything under the Sun!”
Follow Inspire2rise on Twitter. | Follow Inspire2rise on Facebook. | Follow Inspire2rise on YouTube
Discover more from Inspire2Rise
Subscribe to get the latest posts sent to your email.
