Skip to content

Commit 16672b2

Browse files
chbasanichbasani
committed
charitha update
1 parent 4a75eca commit 16672b2

File tree

1 file changed

+133
-15
lines changed

1 file changed

+133
-15
lines changed

‎docs/modules/Module0/intro.md‎

Lines changed: 133 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -1,27 +1,145 @@
1-
---
2-
title: What is Azure OpenAI Service?
3-
parent: Module 0 - Introduction and Pre-requisites
4-
has_children: false
5-
nav_order: 1
6-
---
1+
**What is Azure OpenAI Service?**
72

8-
# What is Azure OpenAI Service?
3+
Azure OpenAI Service is a cloud-based service that provides access to
4+
the OpenAI API. You can use the OpenAI API to perform the following
5+
tasks:
96

10-
Azure OpenAI Service is a cloud-based service that provides access to the OpenAI API. You can use the OpenAI API to perform the following tasks:
7+
- Language Understanding
118

12-
* Generate text
13-
* Perform language understanding
14-
* Perform language translation
9+
- Text Summarization
1510

16-
The OpenAI GPT-3 language model is a large-scale unsupervised language model that can generate coherent text and perform a variety of other language-related tasks.
11+
- Semantic Search
12+
13+
- Conversation AI
14+
15+
- Code Generation
16+
17+
OpenAI is a powerful Language Generative model that predicts the next
18+
token to generate text output based on the input instruction from the
19+
user. Azure OpenAI is the model pretrained and hosted in Azure for
20+
easier deployment for the customer projects.
1721

1822
To learn more about Azure OpenAI Service, you can:
1923

20-
* Check out the [Azure OpenAI Service documentation](https://docs.microsoft.com/en-us/azure/openai/).
24+
- Check out the [Azure OpenAI Service
25+
documentation](https://docs.microsoft.com/en-us/azure/openai/).
2126

22-
## How do I get started with building applications using Azure OpenAI Service?
27+
The user 'Prompt' gives text instructions with the appropriate context.
28+
The more detailed it is with possible examples, it would help the model
29+
to arrive to the right context and generate the result set 'Completion'
30+
that is presented to the user.
31+
32+
You can train the model with one or few-shot examples or with
33+
interactions. The model can be fine-tuned with a few parameters to
34+
customize it to the specific need. The model can be tuned to be
35+
deterministic/probabilistic or instructed to continue with the results
36+
based on these set parameter values.
37+
38+
- GPT-3 is the first offering with the 4 models Ada, Babbage, Curie
39+
and Davinci with the increasing inferencing capabilities, but would
40+
consume more time for presenting the results. The GPT Codex models
41+
supports Co-pilot.
42+
43+
- GPT-35-Turbo is the ChatGPT model option with improved accuracy for
44+
a conversational model.
45+
46+
- GPT-4 is the preview version that allows for a larger token size
47+
prompts and has security built-in. You can request using this
48+
[Access Request
49+
Form](https://customervoice.microsoft.com/Pages/ResponsePage.aspx?id=v4j5cvGGr0GRqy180BHbR7en2Ais5pxKtso_Pz4b1_xURjE4QlhVUERGQ1NXOTlNT0w1NldTWjJCMSQlQCN0PWcu)
50+
51+
## Applications and Use cases:
52+
53+
The language generation from the GPT is based on the semantics of the
54+
Prompt that help it to the inference Completion in the below scenarios
55+
with some examples:
56+
57+
- Writing Assistance:
58+
59+
- Government agency using Azure OpenAI Service to extract and
60+
summarize key information from their extensive library of rural
61+
development reports.
62+
63+
- Financial services using Azure OpenAI Service to summarize
64+
financial reporting for peer risk analysis and customer
65+
conversation summarization.
66+
67+
- Code Generation:
68+
69+
- Aircraft company using to convert natural language to SQL for
70+
aircraft telemetry data.
71+
72+
- Consulting service using Azure OpenAI Service to convert natural
73+
language to query propriety data models.
74+
75+
- Reasoning over data
2376

24-
The best way to get started with building applications using Azure OpenAI Service is to follow the tutorials in this repository.
77+
- Financial services firm using Azure OpenAI Service to improve
78+
search capabilities and the conversational quality of a
79+
customer's Bot experience.
2580

81+
- Insurance companies extract information from volumes of
82+
unstructured data to automate claim handling processes.
2683

84+
- Summarization:
85+
86+
- International insurance company using Azure OpenAI Service to
87+
provide summaries of call center customer support
88+
conversation-logs.
89+
90+
- Global bank uses Azure OpenAI Service to summarize financial
91+
reporting and analyst articles .
92+
93+
## Prompt Engineering:
94+
95+
The model is only as effective as the Prompts sent as input. And this
96+
also trains the models to arrive to a customized model with appropriate
97+
inference context. Here are a few techniques that can support a better
98+
model performance:
99+
100+
1. Structure the input to instruct the model in a step-by-step process
101+
to make it understand the question and suggest it arrive to the
102+
inference.
103+
104+
2. Prompt Chaining helps to elicit more reliable answers and fine tune
105+
it with thousands of Prompts to fine tune it.
106+
107+
3. The models are limited by the Prompt token size for the deployment
108+
type chosen. Long text beyond the token limit is broken into Chunks
109+
and processed.
110+
111+
4. Leverage One-Shot/Few-Shot reasoning to be specific about what is
112+
the expected result set. The model can learn using these scenarios
113+
presented in the Prompt, and you are explicitly telling the mode how
114+
to think by prompting how it should reason for the similar problem.
115+
116+
5. This technique called Chain-of-Thought, is a super powerful
117+
technique, not only can it be used to provide model explainability
118+
(where sometimes GPT-3 is seen as a blackbox) but it can help the
119+
model reason and arrive at a desired output by simply just telling
120+
the model to think step by step.
121+
122+
6. One interesting trick is to have the model decompose the task into
123+
smaller tasks and figure it out on its own. This allows the model to
124+
reason along the way and can lead to much better results. The
125+
technique is called selection-inference prompting.
126+
127+
## Responsible AI (RAI):
128+
129+
The AI models designed for a specific purpose needs to be perceived to
130+
be safe, trustworthy, and ethical. Responsible AI can help proactively
131+
guide these decisions toward more beneficial and equitable outcomes.
132+
133+
- Ensure the model is compliant to the principles of RAI at different
134+
layers of the model deployed with appropriate checks and assessments
135+
at Fine Tuning , at Prompts to generated results , monitoring the
136+
response and the product performance against the expected promises.
137+
138+
- Content Filtering, Feedback channel, Transparency in the product are
139+
a few ways to ensure application is Fair , Reliable , Transparent
140+
and Secure.
141+
142+
## How do I get started with building applications using Azure OpenAI Service?
27143

144+
The best way to get started with building applications using Azure
145+
OpenAI Service is to follow the tutorials in this repository.

0 commit comments

Comments
 (0)