Quick Start with building multilingual classifier by power of OpenAI in 5 minutes

Yoshi Gao included in Software Engineer Notes

2023-04-06 357 words 2 minutes

Contents

image: https://blogs.microsoft.com/blog/2023/01/23/microsoftandopenaiextendpartnership/

What is Openai ChatCompletion

OpenAI Chat API, you can build your own applications with gpt-3.5-turbo and gpt-4

Prompt Engineering Overview

Completion
- Given the unfinished text, complete the text by the GPT model
Zero shot
- Ask the model to generate the text without any prompt example
One shot
- Give one sample before giving the assignment
Few shot
- Give a few samples before giving the assignment
Chain of thought
- Require model to generate the process with inference steps
Role prompting / Instruction Tuning
- Bring in the role to awake model’s functionality

Reference

Pricing

One token is approximately 0.75 word
0.002 USD per 1000 tokens

Reference

https://azure.microsoft.com/zh-tw/pricing/details/cognitive-services/openai-service/

Quick Start with ChatGPT

Installation

1
pip install openai

Create python script

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
import os
import openai
openai.api_type = "azure"
openai.api_version = "2023-03-15-preview" 
openai.api_base = os.getenv("OPENAI_API_BASE")  # Your Azure OpenAI resource's endpoint value.
openai.api_key = os.getenv("OPENAI_API_KEY")

response = openai.ChatCompletion.create(
    engine="gpt-35-turbo", # The deployment name you chose when you deployed the ChatGPT or GPT-4 model.
    messages=[
        {"role": "system", "content": "Assistant is a large language model trained by OpenAI."},
        {"role": "user", "content": "What's the difference between garbanzo beans and chickpeas?"},
    ]
)

print(response)

print(response['choices'][0]['message']['content'])

Reference: https://learn.microsoft.com/en-us/azure/cognitive-services/openai/how-to/chatgpt?pivots=programming-language-chat-completions#working-with-the-chatgpt-and-gpt-4-models-preview

Making the classifier with few-shot learning and instruction tuning

In this example, I will use the gpt-35-turbo model to build a sentiment classifier to classify the text into positive, neutral or negative to inference the person is going to work or not.

I use the following concept:

Few-shot learning
Instruction tuning

	#Note: The openai-python library support for Azure OpenAI is in preview.
	from typing import Dict, List, Text
	from enum import Enum

	from dataclasses import dataclass
	import openai

	openai.api_type = "azure"
	openai.api_version = "2023-03-15-preview"
	openai.api_base = os.getenv("OPENAI_API_BASE") # Your Azure OpenAI resource's endpoint value.
	openai.api_key = os.getenv("OPENAI_API_KEY")


	ENGINE_NAME = "gpt-35-turbo"
	# Instruction tuning to refine the model behaivor
	INSTRUCTIONS = [
	"Classify text as neutral, negative, positive.",
	"Only return neutral, negative or positive."
	]
	# Give the model some examples
	FEW_SHOT_EXAMPLES = [
	{"user": "I'm on the way to work", "assistant": "positive"},
	{"user": "I don't want to work", "assistant": "negative"},
	]


	@dataclass
	class FewShotExample:
	user: Text
	assistant: Text

	def to_openai_message(self) -> List[Dict]:
	return [
	{
	"role": "user",
	"content": self.user,
	},
	{
	"role": "assistant",
	"content": self.assistant,
	}
	]

	class Label(str, Enum):
	POSITIVE = "positive"
	NEGATIVE = "negative"
	NEUTRAL = "neutral"


	def multilingual_classifier(intput_text: Text) -> Label:
	"""Multilingual classifier."""
	label = Label.NEUTRAL

	try:
	few_shot_examples = [FewShotExample(**example) for example in FEW_SHOT_EXAMPLES]
	messages = _get_chat_completion_messages(intput_text, INSTRUCTIONS, few_shot_examples)
	response = openai.ChatCompletion.create(
	engine="gpt-35-turbo", # engine = "deployment_name".
	messages=messages
	)
	label = Label(response['choices'][0]['message']['content'])
	except Exception as e:
	print(e)

	return label


	def _get_chat_completion_messages(
	query_text: Text,
	instructions: List[Text],
	few_shot_examples: List[FewShotExample]
	) -> List[Dict]:
	"""Get messages.

	Official prompt engineering:
	https://learn.microsoft.com/en-us/azure/cognitive-services/openai/how-to/chatgpt?pivots=programming-language-chat-completions#working-with-the-chatgpt-and-gpt-4-models-preview
	"""

	messages = []
	system_content = """Assistant is a multilingual classifier and just give me the output label without any reason.
	Instructions:
	"""
	for instruction in instructions:
	system_content += f"- {instruction}\n"
	messages.append({"role": "system", "content": system_content})
	for few_shot_example in few_shot_examples:
	messages.extend(few_shot_example.to_openai_message())
	messages.append({"role": "user", "content": query_text})

	return messages


	if __name__ == "__main__":
	label = multilingual_classifier("今天有動力上班")
	# Positive
	print(label.value)

	label = multilingual_classifier("I prefer stay in home.")
	# Negative
	print(label.value)

view raw azure_openai_multilingual_classifier.py hosted with ❤ by GitHub

Reference: https://learn.microsoft.com/en-us/azure/cognitive-services/openai/how-to/chatgpt?pivots=programming-language-chat-completions#few-shot-learning-with-chat-completion

FAQ

Q: What’s the difference between openai and azure-openai
A: Azure-openai provide the service monitoring for model usage but they both use openai client library

Q: How to set the request timeout in openai client library?
A: You can set it in the create function

1
2
3
4
5
response = openai.ChatCompletion.create(
    engine=OPENAI_ENGINE_NAME,  # engine = "deployment_name".
    messages=messages,
    request_timeout=3.,
)

Q: Can I specify the api information in the create method?
A: Yes

1
2
3
4
5
response = openai.ChatCompletion.create(
    engine=OPENAI_ENGINE_NAME,  # engine = "deployment_name".
    messages=messages,
    api_version="2023-03-15-preview"
)