A Python package that wraps scweet and gpt-2-simple to allow easy tweet generation based on any existing Twitter account.
The idea of this project is to let anyone unfamiliar with GPT-2 to create their own Twitter bot emulating an existing account (e.g., Deep Leffen, Deep Slacks Bot, Deep Bumpaah Bot) in a single API call. The library handles all boilerplate code and basic hyperparameter tuning involved, allowing the user to spend that time curating tweets they want instead.
twitter_user_emulator can be installed via PyPi:
pip3 install twitter_user_emulator
To enable GPU usage for faster training (recommended), a version of Tensorflow 2.X (min 2.5.1) must be installed on your system.
An example API call:
from twitter_user_emulator import emulator
emulator.emulate_user(user="bumpaah", start_date="2020-01-01", end_date="2021-01-01", out_tweets=2000)
This piece of code will download and train a gpt-2 model to generate 2000 tweets based on all the tweets user @bumpaah made between 2020-01-01 and 2021-01-01.
The model will fail to train if there are not enough tweets as input. Please try to pick a time range with at least 500 tweets.
A notebook tutorial that goes over the underlying logic of this implementation can be found here.
MIT
This repo has no affiliation with OpenAI or Twitter.