Python: 3.7+, current release: 1.0.1 build 2022-10-16 THIS REPOSITORY IS OBSOLETE
This is an obsolete version. Check the newer desktop version called PYGPT.
PYGPT supports GPT4, GPT3, GPT3.5 and DALL-E 2 and offers features like short and long-memory, context storage and restore, editable prompts presets and more. It is available for Windows and Linux (compiled binary desktop versions are available to download from project website).
https://github.com/szczyglis-dev/py-gpt
Enjoy!
<<< END OF INFORMATION, OLD VERSION:
"GPT3-py" is a Bring-Your-Own-Key terminal-based application written in Python 3, allowing easy interaction with the OpenAI's GPT-3 artificial intelligence. The application provides a chat mode in the form of a configurable chatbot, as well as additional functions such as generating code in several different languages (Python, C++, C#, Java, Javascript, TypeScript, PHP, Assembly, SQL, Bash, Ruby, Go, Perl, R, Matlab and Q#), translating the code into a human language, generating Linux and Windows commands based on the description and much more. The application is easy to configure and can be extended with your own features.
INFO: The application requires your own OpenAI's API key to work and is intended for home use on a local machine.
Generative Pre-trained Transformer 3 (GPT-3; stylized GPT·3) is an autoregressive language model that uses deep learning to produce human-like text. Given an initial text as prompt, it will produce text that continues the prompt. From: https://en.wikipedia.org/wiki/GPT-3
- GPT-3 powered chatbot running in the terminal
- defining the terms of conversation with the chatbot, such as time and place
- chatbot configuration
- creating custom queries to GPT-3
- Python, C++, C#, Java, Javascript, TypeScript, PHP, Assembly, SQL, Bash, Ruby, Go, Perl, R, Matlab and Q# code generation from description
- analysis and translation of code operation into human language
- generating commands for Linux from the description
- generating commands for Window (DOS and Powershell) from the description
- easy to configure and modify
- works under any operating system in the terminal
- includes separate scripts that take input from STDIN and command line argument
- built-in output content moderation
- Python: 3.7+
- openai >= 0.22.1
Download or clone repository and install required packages:
pip3 install -r ./requirements.txt
or
pip3 install openai
You must have an OpenAI account and obtain your own API key required for use with the application.
After registering on the https://beta.openai.com/ website, generate your own API key and put it in the config file:
config.py
in section:
API_KEY = "<YOUR API KEY HERE>"
After placing your API key in the config file, the application is ready for use, run it with:
python3 ./gpt3.py
or
./gpt3.py
1 - Chatbot
2 - Custom chatbot
3 - Custom completion
4 - Translate code to human language
5 - Translate into Linux command
6 - Translate into Windows command
7 - Translate into Python code
8 - Translate into PHP code
9 - Translate into Javascript code
10 - Translate into TypeScript
11 - Translate into Java code
12 - Translate into C++ code
13 - Translate into C# code
14 - Translate into SQL code
15 - Translate into Assembly code
16 - Translate into Bash script
17 - Translate into Q# code
18 - Translate into Ruby code
19 - Translate into Go code
20 - Translate into Perl code
21 - Translate into R code
22 - Translate into Matlab code
23 - Translate into Lua code
Default chatbot with which you can talk to GPT-3.
After selecting this option, configuration options will be displayed allowing you to define the terms of the conversation:
SET TEMPERATURE FROM 0.0 to 1.0, WHERE 0.0 = LESS ABSTRACTION, 1.0 = MORE ABSTRACTION, OR LEAVE EMPTY TO LEAVE DEFAULT (0.3):
The above allows to define the temperature of the conversation in the range from 0.0 to 1.0. The lower the value, the more precise chatbot responses are, the higher the value, the more abstract the responses will be.
In the next step, you can define the time of the conversation:
CHOOSE TIME, OR LEAVE EMPTY IF YOU DO NOT WANT TO SET CUSTOM TIME:
0 - Do not set custom time (default)
1 - Custom time
2 - Real time
By default, the time is not specified, but you can enter your own time or choose the real time. When choosing the option to enter your own time, you can enter the full date, time or time of the day, e.g.:
Monday, 2022.10.03, 12:00
or
evening
If you choose real time, the time will be determined automatically according to the current time.
In the last prompt, you can choose the circumstances or place of the conversation, or leave the default selection so as not to specify the place:
CHOOSE PLACE FOR CONVERSATION, OR LEAVE EMPTY IF YOU DO NOT WANT TO SET CUSTOM PLACE:
0 - Do not set custom place (default)
1 - Custom place
Example of choosing a place if you have chosen to define your own place:
[1] and press Enter
> in home
This will result in the following behavior during the conversation:
BEGIN:
------------------------------
> hi, where are we now?
We are currently in your home.
> what do you thing about this place?
I think this place is great! It's really cozy and has a lot of character.
Tip: to end the conversation at any time and exit the application, use the CTRL+C key combination.
In this feature, you can program GPT-3 yourself, so that the conversation proceeds according to the scenario you set.
For help with creating queries, visit the OpenAI website: https://beta.openai.com/docs/guides/completion
Example query:
Batman is a hero. A Human is a person who meets Batman on the street.
Human: Hi
Batman: Hello
Human:<INPUT>
Batman:
Preparing the above scenario will give you the opportunity to "talk to Batman":
BEGIN:
------------------------------
> what are you doing?
I am fighting crime.
Tip: when creating your query, use the placeholder <INPUT> for user input. It will be replaced with the text entered from the terminal.
This feature allows you to prepare your own query for GPT-3, with the difference that the query must be specified during each loop run.
The feature can be used for e.g. data analysis or to generate stories.
Example query:
>>>
Topic: Superman
Story: Superman is a comic superhero who saves people. Once upon a time, Superman had to save the Earth from an alien invasion.
Topic: Tony Stark
Story:
Result:
Tony Stark is a billionaire who creates technology. He is also the superhero Iron Man. Once, Tony Stark had to save the world from a giant robot.
You can find more usage examples on the OpenAI website: https://beta.openai.com/examples
This feature allows any code to be translated into human language. You can enter any code to get an explanation of how it works.
Example query:
> if (a > 1) b = true;
Result:
The code checks if the value of the variable a is greater than 1. If it is, the code sets the value of the variable b to true.
This feature allows you to generate any command for Linux based on the description.
Example query:
> create /home/user/file.txt, copy it to /home/tmp, then zip the files from /home/tmp to files.tar.gz
Result:
cp /home/user/file.txt /home/tmp
tar -czvf files.tar.gz /home/tmp
This feature allows you to generate any command for Windows (DOS + Powershell) based on the description.
Example query:
> create test directory on drive C:, copy files to it from directory D:\files, and finally move files from directory C:\test to drive E:\
Result:
DOS:
md C:\test
copy D:\files C:\test
move C:\test E:\
PowerShell:
New-Item -Path 'C:\test' -ItemType Directory
Copy-Item -Path 'D:\files' -Destination 'C:\test'
Move-Item -Path 'C:\test' -Destination 'E:\'
Features 7 to 23 allow you to generate code based on the description of the action.
The following is an example of use for Python:
> read all files in the ./test directory, pack them all, and save them to packed.zip
Result:
import os
import zipfile
def pack(dir, zip_file):
for root, dirs, files in os.walk(dir):
for file in files:
zip_file.write(os.path.join(root, file))
with zipfile.ZipFile('packed.zip', 'w') as zip_file:
pack('./test', zip_file)
The code generation for other languages is the same.
Here is an example for Bash:
16 - Translate into Bash script
Example query:
if /home/test exists, then copy all files from /home/tmp to it
Result:
if [ -d /home/test ]; then cp -r /home/tmp/* /home/test/; fi
There is a separate script in the repository that takes input from STDIN and returns a single response from GPT-3.
This can be used to connect I/O with other commands and programs.
Example:
python3 ./gpt3_stdin.py
Command in the terminal:
echo "Hello" | python3 ./gpt3_stdin.py
Result:
Hello there! How are you doing today?
There is a separate script in the repository that takes input from the command line argument and returns a single response from GPT-3.
This can be used to connect I/O with other commands and programs.
Example:
python3 ./gpt3_cmd.py
Command in the terminal:
python3 ./gpt3_cmd.py "hello!" > ./output.txt
cat ./output.txt
Result:
Hello! How are you doing today?
The package includes a built-in moderation of responses returned by GPT-3, responsible for filtering inappropriate content.
Moderation can be turned on or off using the parameter:
is_moderation = True|False
located in the file:
config.py
The application has the option of saving the entire history of conversations to .txt files. To enable or disable saving to history use:
store_history = True|False
located in the file:
config.py
By default, history is saved to the ./history directory.
You can freely configure the operation of the application, specifying parameters such as the selected model (the default model is DaVinci), the maximum number of tokens and other options in the config file:
# config.py
is_moderation = False # enable or disable output moderation
store_history = True # enable or disable history storage
store_history_time = True # enable or disable timestamp in history files
history_path = "./history" # path to the history directory
temperature = 0.3 # default temperature
max_tokens = 1000 # max tokens in output
model = "text-davinci-002" # model
top_p = 1 # sampling threshold
frequency_penalty = 0.0 # model’s tendency to repeat predictions
presence_penalty = 0.0 # encourages the model to make novel predictions
# initial values
stop = ""
mode = "chat"
mode_time = 0
mode_place = 0
custom_time = ""
custom_place = ""
date_format = "%Y.%m.%d"
time_format = "%H:%M"
The application is intended for personal home use.
Please note that if you choose to use it for commercial use, please report the use to the App Review procedure on the OpenAI website.
- 1.0.1 - Added additional languages: TypeScript, Ruby, Go, Perl, R, Matlab, Q# (2022-10-16)
- 1.0.0 - Published first release. (2022-10-15)
14X6zSCbkU5wojcXZMgT9a4EnJNcieTrcr
or by PayPal: https://www.paypal.me/szczyglinski
Enjoy!
MIT License | 2022 Marcin 'szczyglis' Szczygliński
https://github.com/szczyglis-dev/gpt3-py
Contact: szczyglis@protonmail.com