# The Complete Manual for the GPT-3 Language Model

Generative GPT-3, also referred to as Pre-trained Transformer 3, is an autoregressive language model developed by OpenAI. It is the biggest language model ever made and was trained using 175 billion parameters and an estimated 45 terabytes of text data! The models’ ability to produce text that resembles that of a human is made possible by their extensive use of data from the internet. It is a deep neural network model for language generation that has been trained to determine the likelihood that a word will appear in a sentence. GPT-3 produces text of such high quality that it can be challenging to tell whether it was written by a human, which has both advantages and disadvantages. In this blog, we have discussed a complete manual for GPT-3 Model and how it can be used.

## Prerequisites

Install OpenAI in colab or jupyter notebook using the” !pip install openai ” command. Get the openai API Key from the beta.openai.com website by registering for it.

## Implementation

We have discussed the Manual for GPT-3 model. Firstly install and import the openai library. It can be done using the below code.

!pip install openai
import os
import openai

We employ the GPT-3 Completion endpoint, which has a number of applications including translation, summarization, question and answer, and others. The following Python codes can be used to perform the general function that accepts texts as input.

def GPT_Completion(texts):
openai.api_key=' Your openai api key here '
response = openai.Completion.create(
engine="text-davinci-002",
prompt =  texts,
temperature = 0.6,
top_p = 1,
max_tokens = 250,
frequency_penalty = 0,
presence_penalty = 0
)
return print(response.choices[0].text)

• The engine is set to the “text-DaVinci-002,” which according to OpenAI’s documentation is the “most capable” GPT-3 model.

• prompt is set to “text,” a variable that represents the text input to the function.

• The model’s output’s temperature indicates how deterministic it is. The model can sample outputs more freely at high temperatures.

• The maximum number of tokens that may be returned is specified by max tokens.

• The parameters frequency penalty and presence penalty both penalize the model for returning outputs that appear frequently.

Different example applications that accept different types of text input can use this function.

## Examples

Example-1:  If we want to write a small blog on a topic, we can use the above function to write it. If we wanted to write on covid-19, use the following code.

# Input

text = ' write a blog on covid-19 with 100 words '
GPT_Completion(text)

The output generated from the GPT-3 Model for the above input is as follows.

Covid-19 is a novel coronavirus first identified in 2019. It is similar to SARS-CoV, the virus that caused the 2002-2004 SARS pandemic. As of June 2019, only a limited number of cases have been identified in people in the Middle East, all of whom have since recovered.

Example 2: You can even ask questions by passing the question as a parameter to the function.

# Input

text = ' Who invented the computer ?'
GPT_Completion(text)

You get the following output for the above input.

The first computer was invented in the early 1800s by Charles Babbage.

Example 3: You can ask to write any code by passing it into the function parameter.

# Input

code = ' Write code for finding odd numbers in python ?'
GPT_Completion(code)

You get the following output for the above input.

odd_numbers = [x for x in range(1,20) if x%2!=0]
print(odd_numbers)

Also, read 176B Parameter Bloom Model.