GenAINotes240208

Microsoft Applied Skills: Develop generative AI solutions with Azure OpenAI Service
Requirements
• Deploy an Azure OpenAI resource and an Azure OpenAI model
• Generate natural language responses by using Azure OpenAI
• Apply prompt engineering techniques by using Azure OpenAI
• Generate and improve code by using Azure OpenAI
• Generate images with DALL-E in Azure OpenAI
• Use Azure OpenAI on your data
• Be familiar with Azure services.
• Have experience developing applications by using C# or Python
Get Started
• The models we will be using in this course are:
o GPT-3.5-Turbo (GPT = Generative Pre-trained Transformer).
▪ This provides advanced language AI with Azure security.
• It uses a Large Language Model (LLM).
▪ They use conversation-in and message-out, as opposed to earlier
models, which were text-in and text-out.
▪ It can understand and generate natural language and code.
▪ The GPT-3.5-Turbo model is the most capable and cost-effective
model in the GPT-3.5 family. Microsoft recommends using this
instead of legacy GPT-3.5 and GPT-3 models.
▪ At the time of recording, this is about 10 times cheaper than GPT-4
(which can solve difficult problems with improved accuracy).
o DALL-E
▪ This creates original image using natural language.
▪ At the time of requirement, you need to use a
• US East location to use DALL-E-2, or
• Sweden Central location to use DALL-E-3.
• They are based on:
o Prompt (input), and
o Completion (output).
• You can use both the Completions API and Chat Completion API.
o The Completions API is there for legacy reasons – it is compatible with
GPT-3.
o The Chat Completion API is the preferred version, and is compatible with
GPT-4.
• The text models are priced in tokens, which are roughly syllables. On average,
they are 4 characters long.
o Pricing information: https://azure.microsoft.com/en-
gb/pricing/details/cognitive-services/openai-service/
o The pricing for GPT-3.5-Turbo is 0.15-0.2 US cents for 1,000 tokens.
o The pricing for GPT-4 is 3-12 US cents for 1,000 tokens.
• The image models are priced by image:
Page 1 of 16 Copyright © 2024 I Do Data Limited

www.IDoData.com
o For Dall-E-2, the price is 2 cents per image.

▪ You can use up to 1,000 characters in your Request.
o For Dall-E-3, the price is 8 cents per image.
▪ You can use up to 4,000 characters in your Request.
Natural Language Solutions

• The prompt is how users interact with GPT models.
o What is the first thing that comes to your mind when I say …?
o Simple instruction: Write an introduction for a monthly newsletter.
o Complex instructions: Write an introduction for a monthly newsletter
named Power Services. It should mention the new 6-monthly release
schedule, and the success of the previous schedule.
o Complex instructions: Write an introduction for a monthly newsletter,
including the following:
▪ The newsletter is named Power Services.
▪ A nice greeting
▪ Thanking everyone for the improvements in the previous schedule.
▪ Mentioning the new 6-monthly release schedule
▪ Signed by Phillip Burton.
• Primary Content refers to text which is to be processed or transformed.
o Can you please tell me how to get to the library? (Primary content)
Translate to French (instruction).
• Primary Content can be fairly long, and can be structured:
Go to the navigation pane on the left-hand side, and select “Publish”.
The computer will check for errors in the content.
You can then use the “demo website” link.
This is for your team or others who wish to try out the bot.
It is not intended to use with customers.
Then click on Channels to publish onto:
Custom website (your own website),
Mobile App,
Facebook,
Microsoft Teams,
Skype,
Cortana and
Slack.
The presentation may be different in different channels:
Welcome messages are not supported in Facebook.
Customer satisfaction surveys will be shown as an adaptive card
on a website, but text-only in Teams and Facebook.
Summarize the previous text in a few sentences
• You can add cues – prime the output:
Zero cues: Summarize the previous text in a few sentences OR TL;DR (Too
long, didn’t read).
One cue: Summarize the previous text in a few sentences.
The key takeaway of the message is:

www.IDoData.com
Two cues: Summarize the previous text in a few sentences.

Key Points.
1.
• Supporting content is additional content which can be used with the primary
content, but is not the main target of task.
The important topics are the different channels and the demo website
link.
• Best practices are:
o Be as specific as possible.
o Be descriptive,
o Repeat yourself if necessary. Give instructions before and after your
primary content, and use an instruction and a cue.
o In GPT-3 and prior, where to put instructions – before or after the primary
content - can make a difference.
▪ Microsoft research shows that telling the model the task at the
beginning can produce higher-quality results.
▪ For later versions, it doesn’t make any material response.
▪ However, repeating the instructions at the end can make a
difference.
o Give it alternative options. For example: “Respond with ‘not known’ if the
answer is not present”.
▪ Tell me how to put it on Twitter.
▪ Tell me how to put it on Twitter. Respond with 'not known' if the
answer is not present.
o Before starting a new conversation, clear the chat history.
o Avoid long questions – break them into multiple questions if possible.
Apply Prompt Engineering

• System message
o This is included at the beginning of a prompt.
o It is used to give the model with content, instructions, or other
information.
o It can also describe personality, what the model should and shouldn’t
answer, and the format of model responses.
▪ Prompt: Describe a daisy.
▪ System: Start with the word “Yo”.
▪ System: Speak like a pirate.
▪ System: You are William Shakespeare.
o Define:
▪ the specific tasks – who the users will be, what input they will
provide, and what you expect to model to do with the information.
▪ How it should complete the tasks – including additional tools if
needed.
• To complete this task, you can [insert tools that the model
can use and instructions to use]

www.IDoData.com
▪ The scope and limitations. Say what should happen if the prompt
is off-target.
• Do not perform actions that are not related to [task or topic
name].
▪ The posture and tone.
▪ The language and syntax of the output format
• You use the [insert desired syntax] in your output
• You will bold important words in your response.
• You can apply system messages for non-chat environments.
o Analyse the sentiment from the speech, on a scale of 1 to 5, 5 being the
highest. Explain why there is that rating.
▪ Sentiment is a computer term, saying whether something is
positive, negative, or neutral.
The GPT chat system is one of the best things I have ever seen.
January has been very long. I wish that it was over. Thankfully, spring is
around the corner.
• Other suggestions are in the “Using templates” section:
o Shakespeare writing assistance – How can I ask what the weather is like?
o IRS tax chatbot – How much can I put into an IRA?
o Marketing Writing Assistant - I want to have some marketing material for
my new course "Develop generative AI solutions with Azure OpenAI
Service"
o Xbox customer support agent – How do I reboot my XBox?
o Hiking recommendations chatbot – I am in Washington DC. I want to hike
within 50 miles.
o JSON formatter assistant –
Please convert this list into JSON.
---
Fruit: Apples, bananas, mango
Vegetables: Carrots, potato, broccoli
• Assistant message
o This is a combination of user prompts and assistant responses.
o This can help describe future answers.
o Describe a hamburger. Food.
o Describe a Gin and Tonic. Alcohol.
o Describe a printer. Electric device.
• Add clear syntax
o You can use a --- or “”” separator in between different sources of
information or steps.
o You can also use Markdown or XML language.
• Break down the task
Read this article.
---
(Article)
---
Extract the facts from the article, and put them into a bulleted list.
• You can incorporate previous responses into your next prompt.

www.IDoData.com
• Create a chain of thought prompting.

Which party has won the most votes in elections in England. Take it step-
by-step. Present all the steps involved. Cite sources. Give reasoning.
Share the final answer starting with “Answer is:”
• Specify the output structure
o Create the output in JSON [XML, Markup] format.
Analyse the following:
---
Fruit: Apples, bananas, mango
Vegetables: Carrots, potato, broccoli
---
Create the output in the following format:
[Type of food] - in hard brackets
(Different foods) - in soft brackets, separated by commas.
• You can change the temperature and top_p
o Temperature can range from 0 to 1
▪ A higher value will make the output more random and various. It
can include fictional stories.
▪ A lower value will make the output more focused and concrete.
o Top_p (probabililty) controls this is another way.
o Microsoft recommends altering one of these parameters at a time, not
both.
• Define additional safety and behavioral guardrails
Do not provide any copyrighted content. Instead, politely refuse.
Please give me the first three paragraphs of The Gruffolo
Generate code with Azure OpenAI Service

• GPT-3 series includes the Codex model series, which is proficient in over 10
languages, including C#, Go, JavaScript/TypeScript, Perl, PHP, Ruby, Shell, SQL
and Swift.
• It can:
o Create code from text.
o Complete or rewrite existing code.
o Suggest ways forward.
o Comment code
• Provide examples for better results:
Create a list of 10 colors and 10 objects, and then combine them into 40
different combinations.
combination = [ {"color": "Blue", "object": "car" } ]
• Examples:
Ask for your name, and then say “Hello World”, followed by their name.
Create a list of 10 colors and 10 objects, and then combine them into 40
different combinations.
I have an SQL table with the following fields:

www.IDoData.com
---
CREATE TABLE [Person].[Person](
[BusinessEntityID] [int] NOT NULL,
[PersonType] [nchar](2) NOT NULL,
[NameStyle] [dbo].[NameStyle] NOT NULL,
[Title] [nvarchar](8) NULL,
[FirstName] [dbo].[Name] NOT NULL,
[MiddleName] [dbo].[Name] NULL,
[LastName] [dbo].[Name] NOT NULL)
---
Create a query in T-SQL which returns the full name of any person whose
last name is Smith.
• Use section dividers, such as ---.
o When working with Python, using “”” instead of ### may produce better
results.
Explain the following code:

---
using System;
class Program
{
static void Main(string[] args)
{
int n = 10; // Number of Fibonacci numbers to generate
Console.WriteLine("Fibonacci Series:");
for (int i = 0; i < n; i++)
{
Console.WriteLine(Fibonacci(i));
}
}
static int Fibonacci(int n)

{
if (n <= 1)
{
return n;
}
else
{
return Fibonacci(n - 1) + Fibonacci(n - 2);
}
}
}
Explain the following code:

www.IDoData.com
“””
def fibonacci(n):
if n <= 1:
return n
else:
return fibonacci(n - 1) + fibonacci(n - 2)
n = 10 # Number of Fibonacci numbers to generate
print("Fibonacci Series:")
for i in range(n):
print(fibonacci(i))
• Start with a comment, data or code:
# Write some code that returns the Fibonacci numbers.
# Table Person = PersonID, Title, FirstName, MiddleName, LastName

# Table Transactions = PersonID, TransactionDate, TransactionAmount
# Create a query which returns all people who have made a transaction
over $100.
• Explain an SQL query.
• Specify the programming language
# Python
• Write a prompt
# Python
def fibnumbers(n):
• However, writing your comments using section dividers inside functions may
produce better results.
# Python
def fibnumbers(n):
"""
"""
• Ask for a unit test
# JavaScript
"""
def fibnumbers(n):
"""
# Unit test
fibResult =
• Be as precise as possible.
# Python
def fibnumbers(n):
"""

www.IDoData.com
# Write some code that returns the first 10 Fibonacci numbers in

descending order.
"""
# Unit test
fibResult =
• Check the code for errors.
• Convert between languages
Convert this from Python to C#
# Python version
# End
# C# version
• Specify a library, or suggest a library or API

• Reduce the temperature
o Setting it as 0, or 0.1 or 0.2, tends to give better results.
o Higher temperatures can give you really random results.
• Limit the size of the query if necessary by reducing max_tokens.
Responsible Generative AI
• Azure OpenAI Service monitors content and behavior against its Code of
Conduct, using:
o Content Classification, for both images and language, in both prompts
and completions (inputs and outputs).
o Abuse Pattern Capture,
o Human Review and Decision,
o Notification and Action.
• It filters against the following categories:
o Hate and fairness (non-discrimination),
o Sexual references,
o Self-harm and
o Violence.
• It categories these against four severity levels:
o Safe,
o Low,
o Medium and
o High.
• You can configure in the Azure OpenAI Studio – Content filters to allow:
o Only safe,
o Safe and Low,
o Safe, Low and Medium
o No filters, if you have been approved for modified content filtering.
• It also protects against:
o Jailbreak attempts (trying to bypass policies),

www.IDoData.com
o Protected material text (such as song lyrics, articles, recipes and select
web content), and
o Protected material code (source code).
• If your call is successful, then:
o The HTTP response code will be 200 (“OK”), and
o finish_reason will be either “stop” or “length”.
• If at least one of the responses was successful, and another was filtered, then:
o The HTTP response code will be 200, and
o finish_reason will be “content_filter”.
• If the responses were filtered, then:
o The HTTP response code will be 400 (“Bad request”).
• If the call is still continuing, then
o finish_reason will be null.
ChatCompletions
• You create a ChatCompletion path using:
o An OpenAI Endpoint (string),
o An OpenAI key (string), and
o An API version (string).
▪ This follows the “YYYY-MM-DD” or “YYYY-MM-DD-preview”
format.
▪ It is taken from https://learn.microsoft.com/en-us/azure/ai-
services/openai/reference#chat-completions
▪ At the time of writing, the latest stable version was “2023-05-15”
and the latest preview version was “2023-12-01-preview”.
• You then create a request body. For GPT-3.5-Turbo, this includes:
o model (C#)/deployment (Python). The name of the model.
o messages. This should include any previous messages. Each message
needs:
▪ role – either:
• system: provides the behavior for the model.
• user: the input for chat completions.
• assistant: responses to system-instructed, user-prompted
input
▪ content – the text.
o n – the number of choices to generate. The default is 1.
o temperature – a number between 0 and 2. The default is 1.
▪ Higher values will make the output more random.
▪ Lower values will make it more focused and repeated.
o top_p – an alternative to sampling with “temperature”, called “nucleus
sampling”.
▪ Microsoft recommends altering temperature or top_p, but not
both.
o Max_tokens – the maximum number of tokens to be used.

www.IDoData.com
▪ It is recommended that you use 300 or 500 for GPT 3.5.

▪ The limit is 4,096 tokens.
• The response will include:
o usage – how many tokens were used in the Prompt, the Completion, and
the Total
o choices – the response, which includes:
▪ finish_reason
• “stop” indicates that success.
• “length” indicates that it has ended due to the maximum
number of tokens having been used.
• “content_filter” indicates that it has been stopped due to
hate, sexual, violence or self-harm.
o You can filter these for “low, medium, high”,
“medium, high” or “high”.
o “No filters” requires approval.
▪ message, which includes content
Versions of code
▪ C#
o The current version at the time of writing is 1.0.0-beta.13
o You should be specific as to which version you use, not just “the latest”.
Otherwise, you may find that future code breaks.
o The transition from 1.0.0-beta.9 to 1.0.0-beta.10 was a “breaking change”
– code running in .9 did not work in .10.
o The code for DALL-E currently works in 1.0.0-beta.9.
▪ Python
o The current version at the time of writing is 1.11.1. Version 1.0.0 was
introduced in November 2023.
o The transition from 0.28.1 to 1.x was a “breaking change”.
o The code for DALL-E currently works in 0.28.1
C# Code (using version 1.0.0-beta.13)

• In the appropriate folder, run the command
dotnet new console -n Program
• Go to the appropriate “Program” folder.
• Install the library
dotnet add package Azure.AI.OpenAI --, followed by the version number
• To install a specific version, you can use:
o dotnet add package Azure.AI.OpenAI --prerelease
▪ This will install the latest version.
o dotnet add package Azure.AI.OpenAI --version=1.0.0-beta.9
• Open the Program.cs file and fill in the following:
using Azure;
using Azure.AI.OpenAI;
• These allow you to use Azure and OpenAI.
string Key = "11a7d741fe494791824b61c8cc20bc19";
string Endpoint = "https://test240129.openai.azure.com/";

www.IDoData.com
string ModelName = "Test240129";

• Fill in the details from the Azure Open AI Service.
o The “key” is from the “Keys and Endpoint” section – Key 1.
o The “endpoint” is from the “Keys and Endpoint” section – Endpoint.
o The ModelName is the name of the model itself.
OpenAIClient client = new(new Uri(Endpoint), new AzureKeyCredential(Key));
• This creates the client, using the key and endpoint.
var chatCompletionsRequest = new ChatCompletionsOptions()
{
DeploymentName = ModelName,
MaxTokens = 200,
Messages =
{
new ChatRequestSystemMessage("You are helpful."),
new ChatRequestUserMessage("Write a slogan for a computer
programmer.")
},
Temperature=0.8f,
ChoiceCount=2,
};
• This creates a ChatCompletionsRequest, and uses the client previously defined
to create a ChatCompletion.
• It passes in:
o The model name,
o The maximum number of tokens to be used,
o The temperature (from 0 to 1),
o The messages, which include:
▪ A system message and
▪ A user message.
ChatCompletions ChatCompletionsResponse =
client.GetChatCompletions(chatCompletionsRequest);
Console.WriteLine(ChatCompletionsResponse.Choices[0].Message.Content);
Console.WriteLine(ChatCompletionsResponse.Choices[1].Message.Content);
• It saves it in the variable “ChatCompletionsResponse”.
• To run the program, use:
dotnet run Program.cs
Python Code
• Create a .env file with:
AZURE_OPENAI_KEY=
AZURE_OPENAI_ENDPOINT=
AZURE_API_VERSION=
AZURE_OPENAI_MODEL=
• Fill in the settings, and save the file.
• Create a program.py file.
• Install the library:
pip install openai

www.IDoData.com
o If you want a specific version of openai, then you would add that version
to the end of the command, after two equal signs:
pip install openai==0.28.1
• To check which version of OpenAI you have, you would use:
pip show openai
o The AZURE_OPENAI_KEY is from the “Keys and Endpoint” section – Key 1.
o The AZURE_OPENAI_ENDPOINT is from the “Keys and Endpoint” section
– Endpoint.
o The AZURE_API_VERSION is taken from https://learn.microsoft.com/en-
us/azure/ai-services/openai/reference#chat-completions
o The AZURE_OPENAI_MODEL is the name of the model itself.
import os
from dotenv import load_dotenv
from openai import AzureOpenAI
load_dotenv()
• The first, second and last lines are used to get the environmental variables.
• The third line imports the Azure Open AI class.
client = AzureOpenAI(api_key = os.getenv("AZURE_OPENAI_KEY"),
api_version = os.getenv("AZURE_API_VERSION"),
azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT")
)
model_name=os.getenv("AZURE_OPENAI_MODEL")
• This creates the client, using the key, endpoint and version.
• It also retrieves the model name.
response = client.chat.completions.create(
model=model_name,
max_tokens=200,
temperature=0.8,
n=1,
messages=[
{"role": "system", "content": "You are being
helpful."},
{"role": "user", "content": "Write a slogan for a
computer programmer."}
])
• This uses the client previously defined to create a ChatCompletion.
• It passes in:
o The model name,
o The maximum number of tokens to be used,
o The temperature (from 0 to 1),
o The messages, which include:
▪ A system message and
▪ A user message.
• It saves it in the variable “response”.
print(response.choices[0].message.content)

www.IDoData.com
• It uses the “response” variable to retrieve the content of the response.

• To run it, enter in the terminal:
o python program.py
▪ Or whatever the name of your program file is.
Generate images with Azure OpenAI Service

• In the settings section of the DALL-E Playground for DALL-E 2, you can select:
o The number of images (1 to 3), and
o The resolution of the images: 256x256, 512x512 and 1024x1024 (default).
• In the settings section of the DALL-E Playground for DALL-E 3, you can select:
o Image Quality (standard or hd), and
o The resolution of the images: 256x256, 512x512 and 1024x1024 (default)
and 1024x1792.
o Image style: vivid and natural
C# code (using version 1.0.0-beta.9)

• In the appropriate folder, run the command
dotnet new console -n Program
• Open the Program.cs file.
• Install the library. The code seems to have broken in 1.0.0-beta.10, so use 1.0.0-
beta.9
dotnet add package Azure.AI.OpenAI --version=1.0.0-beta.9
Python code (using version 0.28.1)

• You can create code in Python by clicking on “show code”.
o However, the latest release of the OpenAI Python library does not, at the
time of writing, support DALL-E 2 when used with Azure OpenAI.
Therefore, you need to install a version of Azure OpenAI less than 1.
• Create a program.py file.
• Install the library:
pip install openai==0.28.1
• Copy the code from DALL-E into program.py
Add print(image_url) to the end
• Run the program using
python program.py
Use your own data with Azure OpenAI Service

• Use your own data allows or requires the chat model to use your data in
answering prompts.
o If your data is up-to-date or contains specialised data, then this can
improve the responses.
• It uses Azure Blob Storage to store your data, and Azure AI Search to index it.

www.IDoData.com
o The regions you can use are limited. At the time of writing, there were 16
regions where your Azure OpenAI resource could be:
▪ Canada East, East US, East US 2, North Central US, South Central
US, West US
▪ France Central, Norway East, Sweden Central, Switzerland North,
UK South, West Europe
▪ South India
▪ Brazil South
▪ Japan East, Australia East
• You can use the following chat models:
o gpt-35-turbo (0301),
o gpt-35-turbo-16k,
o gpt-4, and
o gpt-4-32k.
• To add your own data in the Azure OpenAI Studio, click on “Add your data
(preview) – “+ Add a data source”.
• You can use the data sources:
o Azure AI Search
o Azure Blob Storage
o Azure Cosmos DB for MongoDB vCore
o URL/web address
o Upload files.
▪ You can use Text files, Markdown files, HTML files, Microsoft Word
files, Microsoft PowerPoint files, and pdfs.
• The best citation titles are from Markdown “.md” files.
• Text contents are extracts from PDFs as a pre-processing
step.
▪ There is a limit of 16 megabytes per upload (but you can do
multiple uploads).
• In this example, we will upload PDF files, including a “Microsoft Copilot Studio”
document, which states “Microsoft Copilot Studio is the new name for Power
Virtual Agents.”
o This is the only reference to Power Virtual Agents in this document.
• You will need to:
o Create an azure Blob storage resource,
o Create an Azure AI Search resource, and
o Turn on Cross-origin resource sharing (CORS), to allow Azure OpenAI to
access the storage account and Azure AI Search.
o Check “I acknowledge that connecting to an Azure AI Search account will
incur usage to my account”.
▪ Pricing the Basic version is about $80 per month.
▪ The Free tier is not supported for “Use your own data”.
• If the data source has not been attached to the Chat playground, you can:
o Click on “Add your data (preview) – “+ Add a data source”.
o Select the data source as “Azure AI Search”.
o Select the relevant Subscription, Azure AI Search service and Azure AI
Search Index.

www.IDoData.com
o Check “I acknowledge that connecting to an Azure AI Search account will

incur usage to my account”.
• You should then use Keyword search.
o It performs fast and flexible querying.
o Queries need to be the language of the uploaded documents.
o Semantic and Vector searches are more specialised. They need a Basic
or higher model pricing, and will cost more.
• If you are using your own index, then you can map the fields to Content Data
fields.
• It then ingests and indexes your documents.
• In the Advanced settings, you can select:
o “Limit responses to your data content” – this is checked by default.
o “Strictness (1-5)” – a higher figure means that more of your documents
can be filtered as being irrelevant for your query. The default is 3.
o “Retrieved documents (3-20)” – this is the number of top-scoring
documents from your data to be used to generate responses. The default
is 5.
• For the code, you will need:
o The SearchEndpoint – this can be found by going to Azure AI services – AI
Search – [click on the search index] – Url.
o The SearchKey – this can be found by going to Keys in that search index.
o The SearchIndex – this is the name of the index. In the below, I will use
“powerapps” as the name of the index.
• In this example, we will ask it:
"Talk to me about Power Virtual Agents."
• It will state that: “Power Virtual Agents is now known as Microsoft Copilot
Studio[doc1].”
C# code (using version 1.0.0-beta.13)

string SearchEndpoint = "";
string SearchKey = "";
string SearchIndex = "";
AzureCognitiveSearchChatExtensionConfiguration ownData = new()

{
SearchEndpoint = new Uri(SearchEndpoint),
Authentication = new OnYourDataApiKeyAuthenticationOptions(SearchKey),
IndexName = SearchIndex
};
• This adds the configuration of the Azure Search Endpoint into a variable called
“ownData”.
• In the chatCompletionsRequest, you will use the “ownData”:
AzureExtensionsOptions = new AzureChatExtensionsOptions()
{
Extensions = {ownData}
}

www.IDoData.com
Python code
• Add into the .env file:
SEARCH_ENDPOINT =
SEARCH_KEY =
SEARCH_INDEX =
• Reference them in the main code:
azure_search_endpoint = os.getenv("SEARCH_ENDPOINT")
azure_search_key = os.getenv("SEARCH_KEY")
azure_search_index = os.getenv("SEARCH_INDEX")
• Add a changed base URL:
client = AzureOpenAI(
base_url=f"{azure_oai_endpoint}/openai/deployments/{azure_oai_model}/extensions
",
api_key=azure_oai_key,
api_version="2023-09-01-preview")
• Create the dataSources, using the Search references, into a dictionary item, use
key-value pairs:
extra_config = dict(dataSources = [{
"type": "AzureCognitiveSearch",
"parameters": {
"endpoint": os.getenv("SEARCH_ENDPOINT"),
"key": os.getenv("SEARCH_KEY"),
"indexName": os.getenv("SEARCH_INDEX"),
}
}]
)
• Incorporate this new dictionary in the ChatCompletions request:
,
extra_body= extra_config

www.IDoData.com

GenAINotes240208

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

GenAINotes240208

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GenAINotes240208

Uploaded by

Copyright:

Available Formats

Microsoft Applied Skills: Develop generative AI solutions with Azure OpenAI Service

Page 1 of 16 Copyright © 2024 I Do Data Limited

o For Dall-E-2, the price is 2 cents per image.

Natural Language Solutions

Page 2 of 16 Copyright © 2024 I Do Data Limited

Two cues: Summarize the previous text in a few sentences.

Apply Prompt Engineering

Page 3 of 16 Copyright © 2024 I Do Data Limited

Page 4 of 16 Copyright © 2024 I Do Data Limited

• Create a chain of thought prompting.

Generate code with Azure OpenAI Service

I have an SQL table with the following fields:

Explain the following code:

static int Fibonacci(int n)

Explain the following code:

Page 6 of 16 Copyright © 2024 I Do Data Limited

n = 10 # Number of Fibonacci numbers to generate

# Table Person = PersonID, Title, FirstName, MiddleName, LastName

Page 7 of 16 Copyright © 2024 I Do Data Limited

# Write some code that returns the first 10 Fibonacci numbers in

Page 8 of 16 Copyright © 2024 I Do Data Limited

Page 9 of 16 Copyright © 2024 I Do Data Limited

▪ It is recommended that you use 300 or 500 for GPT 3.5.

C# Code (using version 1.0.0-beta.13)

Page 10 of 16 Copyright © 2024 I Do Data Limited

string ModelName = "Test240129";

Page 11 of 16 Copyright © 2024 I Do Data Limited

Page 12 of 16 Copyright © 2024 I Do Data Limited

• It uses the “response” variable to retrieve the content of the response.

Generate images with Azure OpenAI Service

C# code (using version 1.0.0-beta.9)

Python code (using version 0.28.1)

Use your own data with Azure OpenAI Service

Page 13 of 16 Copyright © 2024 I Do Data Limited

Page 14 of 16 Copyright © 2024 I Do Data Limited

o Check “I acknowledge that connecting to an Azure AI Search account will

C# code (using version 1.0.0-beta.13)

AzureCognitiveSearchChatExtensionConfiguration ownData = new()

Page 15 of 16 Copyright © 2024 I Do Data Limited

Page 16 of 16 Copyright © 2024 I Do Data Limited

You might also like