Microsoft Phi-3, a family of open AI models that are highly capable and cost-effective small language models (SLMs) has recently been introduced. The Phi-3 mini is a 3.8B language model and is the first model in its class to support a context window of up to 128K tokens with little impact on quality. It is an instruction-tuned model trained to follow different instructions reflecting how people normally communicate.
The model is available on Azure AI to take advantage of the deploy-eval-finetune toolchain and on Ollama for developers to run locally on their laptops. It has been optimized for ONNX Runtime with support for Windows DirectML along with cross-platform support across graphics processing unit (GPU), CPU, and even mobile hardware. It is also available as an NVIDIA NIM microservice with a standard API interface that can be deployed anywhere.
In the upcoming weeks, additional models will be added to the Microsoft Phi-3 family to offer customers even more flexibility across the quality-cost curve. Phi-3 small (7B) and Phi-3 medium (14B) will be available in the Azure AI model catalog and other models shortly.
Microsoft is putting all its efforts into offering the best models across the quality-cost curve and the release of Phi-3 has expanded the selection of models. In essence, Phi-3 stands out for its performance, cost-effectiveness, ease of deployment, and adaptability across different hardware platforms, making it a promising addition to AI language models.
Limitations of Microsoft Phi-3
- Phi-3 mini has a limited context window. Its default context length is 4K which might not be sufficient enough for certain applications.
- Although Phi-3 models can extend the context length to 128K, the performance impact of this extension is not explicitly stated.
- Certain search results indicate that Phi-3 models have not been extensively benchmarked against other models like Llama. No specific information is there on Phi-3's performance on certain tasks such as code generation, reasoning, etc.
- Smaller AI models like Phi-3 are likely to reduce the environmental footprint of AI, but there are no direct limitations of the model itself.
- The model is currently available on Azure AI, and Ollama but its accessibility might be limited.
- There's a hardware requirement for Phi-3 models to perform optimally as Phi-3 models are designed to run on consumer GPU or AI-acceleration hardware.
- While Phi-3 models are designed to be customizable, the extent of this customization and the resources required to do so are not explicitly stated.
Also Read:
Microsoft Invests $1.5 Billion in UAE-based G42 to Propel AI Development
Coca-Cola and Microsoft Sign $1.1Bn Deal to Transform Business with AI
Cognizant & Microsoft Join Hands to Expand Adoption of Gen AI