#phi3

Introducing new generative AI models, Phi 3 fine tuning, and other Azure AI enhancements to enable businesses to scale and personalise AI applications.

All sectors are being transformed by artificial intelligence, which also creates fresh growth and innovation opportunities. But developing and deploying artificial intelligence applications at scale requires a reliable and flexible platform capable of handling the complex and varied needs of modern companies and allowing them to construct solutions grounded on their organisational data. They are happy to share the following enhancements to enable developers to use the Azure AI toolchain to swiftly and more freely construct customised AI solutions:

Developers can rapidly and simply customise the Phi-3-mini and Phi-3-medium models for cloud and edge scenarios with serverless fine-tuning, eliminating the need to schedule computing.

Updates to Phi-3-mini allow developers to create with a more performant model without incurring additional costs. These updates include a considerable improvement in core quality, instruction-following, and organised output.

This month, OpenAI (GPT-4o small), Meta (Llama 3.1 405B), and Mistral (Large 2) shipped their newest models to Azure AI on the same day, giving clients more options and flexibility.

Value unlocking via customised and innovative models

Microsoft unveiled the Microsoft Phi-3 line of compact, open models in April. Compared to models of the same size and the next level up, Phi-3 models are their most powerful and economical small language models (SLMs). Phi 3 Fine tuning a tiny model is a wonderful alternative without losing efficiency, as developers attempt to customise AI systems to match unique business objectives and increase the quality of responses. Developers may now use their data to fine-tune Phi-3-mini and Phi-3-medium, enabling them to create AI experiences that are more affordable, safe, and relevant to their users.

Phi-3 models are well suited for fine-tuning to improve base model performance across a variety of scenarios, such as learning a new skill or task (e.g., tutoring) or improving consistency and quality of the response (e.g., tone or style of responses in chat/Q&A). This is because of their small compute footprint and compatibility with clouds and edges. Phi-3 is already being modified for new use cases.

Microsoft and Khan Academy are collaborating to enhance resources for educators and learners worldwide. As part of the partnership, Khan Academy is experimenting with Phi-3 to enhance math tutoring and leverages Azure OpenAI Service to power Khanmigo for Teachers, a pilot AI-powered teaching assistant for educators in 44 countries. A study from Khan Academy, which includes benchmarks from an improved version of Phi-3, shows how various AI models perform when assessing mathematical accuracy in tutoring scenarios.

According to preliminary data, Phi-3 fared better than the majority of other top generative AI models at identifying and fixing mathematical errors made by students.

Additionally, they have optimised Phi-3 for the gadget. To provide developers with a strong, reliable foundation for creating apps with safe, secure AI experiences, they launched Phi Silica in June. Built specifically for the NPUs in Copilot+ PCs, Phi Silica expands upon the Phi family of models. The state-of-the-art short language model (SLM) for the Neural Processing Unit (NPU) and shipping inbox is exclusive to Microsoft Windows.

Today, you may test Phi 3 fine tuning in Azure AI

Azure AI’s Models-as-a-Service (serverless endpoint) feature is now widely accessible. Additionally, developers can now rapidly and simply begin developing AI applications without having to worry about managing underlying infrastructure thanks to the availability of Phi-3-small via a serverless endpoint.

The multi-modal Phi-3 model, Phi-3-vision, was unveiled at Microsoft Build and may be accessed via the Azure AI model catalogue. It will also soon be accessible through a serverless endpoint. While Phi-3-vision (4.2B parameter) has also been optimised for chart and diagram interpretation and may be used to produce insights and answer queries, Phi-3-small (7B parameter) is offered in two context lengths, 128K and 8K.

The community’s response to Phi-3 is excellent. Last month, they launched an update for Phi-3-mini that significantly enhances the core quality and training after. After the model was retrained, support for structured output and instruction following significantly improved.They also added support for |system|> prompts, enhanced reasoning capability, and enhanced the quality of multi-turn conversations.

They also keep enhancing the safety of Phi-3. In order to increase the safety of the Phi-3 models, Microsoft used an iterative “break-fix” strategy that included vulnerability identification, red teaming, and several iterations of testing and improvement. This approach was recently highlighted in a research study. By using this strategy, harmful content was reduced by 75% and the models performed better on responsible AI benchmarks.

Increasing model selection; around 1600 models are already accessible in Azure AI They’re dedicated to providing the widest range of open and frontier models together with cutting-edge tooling through Azure AI in order to assist clients in meeting their specific cost, latency, and design requirements. Since the debut of the Azure AI model catalogue last year, over 1,600 models from providers such as AI21, Cohere, Databricks, Hugging Face, Meta, Mistral, Microsoft Research, OpenAI, Snowflake, Stability AI, and others have been added, giving us the widest collection to date. This month, they added Mistral Large 2, Meta Llama 3.1 405B, and OpenAI’s GPT-4o small via Azure OpenAI Service.

Keeping up the good work, they are happy to announce that Cohere Rerank is now accessible on Azure. Using Azure to access Cohere’s enterprise-ready language models Businesses can easily, consistently, and securely integrate state-of-the-art semantic search technology into their applications because to AI’s strong infrastructure. With the help of this integration, users may provide better search results in production by utilising the scalability and flexibility of Azure in conjunction with the highly effective and performant language models from Cohere.

With Cohere Rerank, Atomicwork, a digital workplace experience platform and a seasoned Azure user, has greatly improved its IT service management platform. Atomicwork has enhanced search relevancy and accuracy by incorporating the model into Atom AI, their AI digital assistant, hence offering quicker, more accurate responses to intricate IT help enquiries. Enterprise-wide productivity has increased as a result of this integration, which has simplified IT processes.

SLMs

Azure are pleased to present the Phi 3 family of open AI models, which was created by Microsoft. Phi 3 models outperform models of the same size and next size up across a variety of language, reasoning, coding, and math benchmarks, making them the most capable and economical small language models (SLMs) on the market. With this update, users can now choose from a wider range of high-quality models, providing them with more useful options for creating and developing generative AI applications.

Phi 3

Phi-3-mini, a 3.8B language model, is now accessible on Hugging Face, Ollama, and Microsoft Azure AI Studio.

There are two context-length variations of Phi-3-mini: 4K and 128K tokens. With little effects on quality, it is the first model in its class to provide a context window of up to 128K tokens.

It is instruction-tuned, which means that it has been taught to comply with various directives that mimic human speech patterns. This guarantees that the model is operational right out of the box.

To utilise the deploy-eval-finetune toolchain, it is available on Azure AI. Alternatively, developers can execute it locally on their computers using Ollama.

With support for Windows DirectML and cross-platform compatibility for CPU, graphics processing unit (GPU), and even mobile hardware, it has been tuned for ONNX Runtime.

It can also be deployed anywhere as an NVIDIA NIM microservice with a conventional API interface. and has been NVIDIA GPU optimised.

Phi language model

The Phi 3 family will see the addition of new variants in the upcoming weeks to provide customers with even greater flexibility across the quality-cost curve. Soon, Phi-3-small (7B) and Phi-3-medium (14B) will be accessible in various model gardens and the Azure AI model catalogue.

Microsoft is still providing the top models across the quality-cost curve, and the Phi-3 release today adds cutting-edge tiny models to the lineup.

Revolutionary performance in a compact package

On important benchmarks, Phi 3 models perform much better than language models of the same and bigger sizes (see benchmark numbers below, higher is better). Phi-3-medium and Phi-3-small perform better than much larger models, such as GPT-3.5T, and Phi-3-mini outperforms models twice its size.

To guarantee comparability, the same pipeline is used to produce all reported numbers. These figures may therefore vary slightly from other published figures as a result of variations in the assessment process. Their technical document contains more information about benchmarks.

Note: Because Phi 3 models have a smaller model size, their ability to recall facts is reduced, which explains why they perform worse on factual knowledge benchmarks (like TriviaQA).image credited to Azure

Microsoft Phi Model

The Microsoft Responsible AI Standard, a set of guidelines applicable to the entire organization and founded on six guiding principles accountability, transparency, fairness, reliability and safety, privacy and security, and inclusivity was followed in the development of Phi-3 models. To assist guarantee that Phi 3 models are responsibly built, tested, and deployed in accordance with Microsoft’s standards and best practices, these models undergo stringent safety measurement and evaluation, red-teaming, sensitive usage review, and adherence to security recommendations.

Enhancing their previous Phi model work (“Textbooks Are All You Need”), Phi-3 models are trained with high-quality data likewise. Extensive safety post-training was implemented to further improve them, including automated testing and evaluations across dozens of harm categories, manual red-teaming, and reinforcement learning from human feedback (RLHF). their technical paper describes their approach to safety training and evaluations, and the model cards include a list of suggested uses and limits. View the sample card set.

Gaining access to new abilities

Because of Microsoft’s experience delivering copilots and helping clients use Azure AI to use generative AI to transform their businesses, the requirement for different-size models across the quality-cost curve for various jobs is becoming more and more necessary. Phi 3 and other little language models are particularly useful for:

Settings with limited resources, such as offline and on-device inference situations.

Situations with latency constraints where quick reactions are essential.

Use cases with less resources, especially those involving easier activities.

See their Microsoft Source Blog for more information on small language models.

Phi-3 models can be applied in inference situations with limited computing power because of their reduced size. Specifically, Phi-3-mini can be utilized on-device, especially after being further optimized for cross-platform availability using ONNX Runtime. Phi-3 models’ reduced size also makes fine-tuning or customization simpler and less expensive. They are also a less expensive solution with significantly improved latency due to their reduced processing requirements. Taking in and deciphering vast text content documents, web pages, code, and more is made possible by the longer context window. Phi-3-mini is a strong contender for analytical work because of its high logic and reasoning ability.

With Phi 3, customers are already developing solutions. Agriculture is one industry where Phi 3 is already proving its worth, as internet access may not always be convenient. Farmers can now use powerful tiny models like Phi-3 and Microsoft copilot templates at any time, along with the added benefit of lower operating costs, further democratizing AI technologies.

As part of their ongoing partnership with Microsoft, ITC, a well-known Indian corporate conglomerate, is utilizing Phi 3 on the copilot for Krishi Mitra, an app that targets farmers and has over a million users.

Phi Model

Phi models were first developed by Microsoft Research and have since been widely utilized; Phi-2 has been downloaded more than 2 million times. The Phi series of models has produced impressive results through creative scaling and thoughtful data curation. A model for Python coding called Phi-1 is followed by Phi-1.5, which improves reasoning and comprehension, and Phi-2, a 2.7 billion-parameter model that outperforms models up to 25 times its size in language comprehension. Every iteration has pushed the boundaries of traditional scaling rules by utilizing superior training data and knowledge transfer strategies.