MODEL, OUTPUT_DIR, TRAIN_STEPS, TITLE = parse_arguments()
#For Mark's machine
os.environ["CUDA_VISIBLE_DEVICES"] = "1,2"
if not os.path.isdir(os.path.join("models", MODEL)):
print(f"Downloading {MODEL} model...")
gpt2.download_gpt2(
model_name=MODEL
) # model is saved into current directory under /models/124M/
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = gpt2.start_tf_sess()
#gpt2.load_gpt2(sess, model_name = MODEL)
# Train Topic generator
print("TRAINING TOPIC GENERATOR...")
if TRAIN_STEPS > 0:
gpt2.finetune(sess,
dataset=TRAIN_TOPIC_PATH,
model_name=MODEL,
steps=TRAIN_STEPS,
checkpoint_dir="topic_gen_chkpts",
multi_gpu=True,
val_dataset=VALID_TOPIC_PATH,
val_every=10)
#gpt2.load_gpt2(sess, model_name = MODEL)
from gpt_2 import finetune, start_tf_sess
import argparse
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description='TensorFlow Haiku GPT-2 Finetuned Language Model')
parser.add_argument('--dataset',
type=str,
help='location of the data corpus')
parser.add_argument('--model_name',
type=str,
default='models/117M',
help='Pretrained model path')
parser.add_argument('--steps',
type=int,
default=1000,
help='No of epochs to train during finetuning')
args = parser.parse_args()
sess = start_tf_sess()
finetune(sess,
dataset=args.dataset,
model_name=args.model_name,
steps=args.steps,
restore_from='fresh',
run_name='run1',
print_every=100,
sample_every=500,
save_every=500)
print("Finetuning completed!!!")
import asyncio
import discord
from discord.ext import commands
import os, datetime, numpy as np, json, re, sys, time
import gpt_2, tensorflow as tf
import re
sess = gpt_2.start_tf_sess()
print(gpt_2.load_gpt2(sess))
checkpoint_path = os.path.join('models', '345M')
enc = gpt_2.encoder.get_encoder(checkpoint_path)
hparams = gpt_2.model.default_hparams()
with open(os.path.join(checkpoint_path, 'hparams.json')) as f:
hparams.override_from_dict(json.load(f))
context = tf.compat.v1.placeholder(tf.int32, [1, None])
bot = commands.Bot(command_prefix='none')
f = open("bot.txt")
bot_token = f.read()
emojis = dict()
imitates = dict()
session = dict()
messages = dict()
messagequeue = dict()
tf_sample = gpt_2.sample.sample_sequence(hparams=hparams,
length=728,