Scribd Logo
Upload

Welcome to Scribd!

  • 12 views

    AI 050T00A ENU PowerPoint 01

    Uploaded by

    wciscato

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd

    AI 050T00A ENU PowerPoint 01

    Uploaded by

    wciscato
    12 views17 pages

    Original Title

    AI-050T00A-ENU-PowerPoint-01

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    12 views17 pages

    AI 050T00A ENU PowerPoint 01

    Uploaded by

    wciscato

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    of 17

    AI-050: Develop

    Generative AI
    Solutions with Azure
    OpenAI Service

    © Copyright Microsoft Corporation. All rights reserved.


    Get started with Azure
    OpenAI Service

    © Copyright Microsoft Corporation. All rights reserved.


    Agenda

    • What is generative AI?

    • Provision a resource

    • Deploying a model

    • Using Azure OpenAI studio

    © Copyright Microsoft Corporation. All rights reserved.


    Learning Objectives
    After completing this module, you will be able to:

    1 Describe what generative AI is

    2 Provision Azure OpenAI resources and deploy OpenAI


    models

    3 Use generative AI to get completions from models in the Azure OpenAI


    studio
    What is generative AI?

    1950 Artificial Intelligence


    s the field of computer science that seeks to create
    Artificial Intelligence intelligent machines that can replicate or exceed
    human intelligence

    1990
    Machine Learning s
    Machine Learning
    subset of AI that enables machines to learn from
    existing data and improve upon that data to make
    decisions or predictions
    Deep Learning 2010
    s Deep Learning
    a machine learning technique in which layers of
    neural networks are used to process data and
    make decisions
    2021 Generative AI
    Generative AI Create new written, visual, and auditory content
    given prompts or existing data.

    © Copyright Microsoft Corporation. All rights reserved.


    Provision an Azure OpenAI resource in Azure

    Deploy a model in Azure OpenAI Studio to use it

    1. Apply for access to the Azure OpenAI service:


    https://aka.ms/oaiapply
    2. Create an Azure OpenAI resource in the
    Azure portal
    Alternatively, use the Azure CLI
    az cognitiveservices account create \
    -n MyOpenAIResource \
    -g MyResourceGroup \
    -l eastus \
    --kind OpenAI \
    --sku s0 \
    --subscription subscriptionID

    © Copyright Microsoft Corporation. All rights reserved.


    Azure OpenAI Studio

    • Web portal for working with Azure OpenAI


    models:
    https://oai.azure.com/
    • View and deploy base models
    • Connect your own data source
    • Manage fine tuning and data files for
    custom models
    • Test models in visual playgrounds:
    • Chat (GPT-3.5-Turbo and later
    models)
    • Completions (GPT-3 and earlier
    models)
    • DALL-E (Image generations)
    • Assistants (Custom and Copilot-like
    experiences)

    © Copyright Microsoft Corporation. All rights reserved.


    Types of generative AI model

    Model Family Description

    GPT-4 Newest, most capable chat-


    based models for language and
    code generation

    GPT-3.5 Natural language and code-


    generation models

    Embeddings Models that use embeddings for


    specific tasks (similarity, text
    search, and code search)

    DALL-E Image-generation model

    © Copyright Microsoft Corporation. All rights reserved.


    Deploying generative AI models

    Deploy a model in Azure OpenAI Studio to use it

    • You can deploy one or more instances of each


    available model
    • The number of deployments depends on your
    quota, which you can see in the portal
    • Alternatively, use the Azure CLI

    az cognitiveservices account deployment create \


    -g myResourceGroupName \
    -n MyOpenAIResource \
    --deployment-name my-gpt-model \
    --model-name gpt-35-turbo \
    --model-version “0301" \
    --model-format OpenAI \
    --scale-settings-scale-type "Standard"

    © Copyright Microsoft Corporation. All rights reserved.


    Using prompts to get completions from models
    Task Prompt Completion
    Classifying content Tweet: I enjoyed the training Positive
    course.
    Sentiment:
    Generating new content Write a poem about databases Databases, oh databases,
    You keep our information safe,
    From the small to the large,
    You store our data in a place.
    Transformation/Translation English: Hello Bonjour
    French:
    Summarization Scotland is [long description of Scotland is [summarized description…]
    Scotland…]

    Summarize the previous text


    Continuation One way to grow tomatoes is to start with seeds…
    Question answering How many moons does Earth have? Earth has one moon.
    Chat Setup, followed by messages… A sequence of relevant responses

    © Copyright Microsoft Corporation. All rights reserved.


    Testing models in Azure OpenAI Studio playground

    © Copyright Microsoft Corporation. All rights reserved.


    Model parameters and tokens in Azure OpenAI

    • Control model behavior through parameters in


    studio pane or API call
    • Parameters include:
    • Max Response: Limit on the number of
    tokens the response can include
    • Temperature: Controls randomness, with a
    higher number producing a less
    deterministic response
    • Top P: Controls randomness similarly but in
    a different way than temperature. If
    adjusting these two values, try one or the
    other but not both
    • Frequency penalty: New tokens are
    penalized for their existing frequency in the
    text so far, reducing the likelihood to repeat
    the same line
    • Presence penalty: New tokens are
    penalized whether they appear in the text so
    far, increasing likelihood of talking about
    new topics
    • Tokens are text measurements, roughly four
    English characters long. Tokens are used for
    measuring model capacity, quotas,© and prompt
    Copyright Microsoft Corporation. All rights reserved.
    Exercise: Get started with Azure OpenAI Service

    Use the hosted lab environment if provided, or view


    the lab instructions at the link below:

    https://aka.ms/mslearn-get-started-azure-openai

    © Copyright Microsoft Corporation. All rights reserved.


    Knowledge check

    What do you need in order to test a generative AI model using the Azure OpenAI Service Studio?
    1  A deployed model name and Azure command line interface
     An Azure OpenAI resource and an Azure Cognitive Services resource
     An Azure OpenAI resource, a deployed model, and a playground

    2 Which parameter could you adjust to change the randomness or creativeness of completions
     Temperature
     Frequency penalty
     Stop sequence

    3 Which Azure OpenAI Studio playground is able to support conversational interactions that
    consist of a sequence of messages?
     Completions
     Chat
     Bot

    © Copyright Microsoft Corporation. All rights reserved.


    Learning Recap

    In this module, we:

    Introduced generative AI and how it relates to AI and machine learning


    Provisioned Azure OpenAI resources
    Deployed OpenAI models
    Generated completions in Azure OpenAI studio

    © Copyright Microsoft Corporation. All rights reserved.


    Resources

    Get started with Azure OpenAI Service

    https://aka.ms/mslearn-start-azure-openai

    © Copyright Microsoft Corporation. All rights reserved.


    Thank you.

    © Copyright Microsoft Corporation. All rights reserved.

    You might also like