ebooks/bot.py

76 lines
2.1 KiB
Python

from argparse import ArgumentParser
from random import randint, choice
from mastodon import Mastodon
from transformers import AutoTokenizer, AutoModelForCausalLM
parser = ArgumentParser()
parser.add_argument('-i', '--instance', help='Mastodon instance hosting the bot')
parser.add_argument('-t', '--token', help='Mastodon application access token')
parser.add_argument('-n', '--input', help='initial input text')
parser.add_argument('-d', '--data', default='data',
help='data for automatic input generation')
parser.add_argument('-m', '--model', default='model',
help='path to load saved model')
args = parser.parse_args()
tokenizer = AutoTokenizer.from_pretrained('distilgpt2')
model = AutoModelForCausalLM.from_pretrained(args.model)
if args.input is None:
# Create random input
if randint(0, 1) == 0:
args.input = choice([
'I am',
'My life is',
'Computers are',
'This is',
'My',
'I\'ve',
'No one',
'I love',
'I will die of',
'I',
'The',
'Anime',
'I\'m going to die',
'Hello',
'@ta180m@exozy.me',
'Life',
'My favorite',
'I\'m not',
'I hate',
'I think',
'In my opinion',
'Breaking news:',
'Have I ever told you that',
'I read on the news that',
'I never knew that'
])
else:
with open(args.data, 'r') as f:
line = choice(f.readlines()).split()
args.input = line[0] + ' ' + line[1]
# Run the input through the model
print(args.input)
inputs = tokenizer.encode(args.input, return_tensors="pt")
output = tokenizer.decode(model.generate(
inputs, do_sample=True, max_length=150, top_p=0.9)[0])
print(output)
# Post it to Mastodon
mastodon = Mastodon(
access_token=args.token,
api_base_url=args.instance
)
post = output.split('\n')[0]
if len(post) < 200:
post = output.split('\n')[0] + '\n' + output.split('\n')[1]
mastodon.status_post(post[:500])