Files
Clover-Edition/generator/ctrl/ctrl_generator.py
T

321 lines
13 KiB
Python

import tensorflow as tf
import numpy as np
tf.enable_eager_execution()
import generator.ctrl.model.transformer as transformer
import re
from collections import Counter
from tensorflow.python import debug as tf_debug
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import embedding_ops
import fastBPE
from story.utils import *
import warnings
warnings.filterwarnings("ignore")
# the loss function is a simple categorical crossentropy between the logits and the labels
def loss(labels, logits):
return tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)
class CTRLGenerator():
def __init__(self, control_code="Apocalypse ", generate_num=28, temperature=0.5, topk=40, nucleus_prob=0):
self.generate_num=generate_num
model_dir = "generator/ctrl/training_utils/seqlen256_v1.ckpt/"
self.control_code = control_code
vocab_file = 'generator/ctrl/model/vocab'
code_file = 'generator/ctrl/model/codes'
self.max_new_lines = 5
# load the vocabulary from file
vocab = open(vocab_file, encoding='utf-8').read().split('\n')
vocab = list(map(lambda x: x.split(' ')[0], vocab)) + ['<unk>'] + ['\n']
print('{} unique words'.format(len(vocab)))
# length of the vocabulary
vocab_size = len(vocab)
# define the numericalization map
# idx2word maps the numericalized ID to the word
# word2idx maps the word to the numericalized ID
self.word2idx = {u: i for i, u in enumerate(vocab)}
self.idx2word = np.array(vocab)
# sequence length to use for the transformer
# the model is trained with a self.seq_length of 512
# so, any value <= 512 should work
self.seq_length = min(self.generate_num, 256)
# the dimension of the transformer
embedding_dim = 1280
# input for the keras model
tokens = tf.keras.layers.Input(shape=(self.seq_length,), dtype='int32')
# Now, we begin defining the model
# we defer the transformer definition to transformer.py
# here, we only define the tied softmax layer
# this layer ties the softmax weights to the input embeddings
class TiedEmbeddingSoftmax(tf.keras.layers.Layer):
def __init__(self, vocab_size=vocab_size, embedding_size=embedding_dim, **kwargs):
super(TiedEmbeddingSoftmax, self).__init__()
self.w = self.add_weight(name='w', shape=(vocab_size, embedding_size),
initializer='random_normal',
trainable=True)
self.b = self.add_weight(name='b', shape=(vocab_size,),
initializer='zeros',
trainable=True)
def call(self, inputs, embed=True):
if embed:
dtype = tf.keras.backend.dtype(inputs)
if dtype != 'int32' and dtype != 'int64':
inputs = math_ops.cast(inputs, 'int32')
return embedding_ops.embedding_lookup(self.w, inputs)
else:
return tf.tensordot(inputs, tf.transpose(self.w), 1) + self.b
# instantiates a tied softmax class
tied_embedding_softmax = TiedEmbeddingSoftmax()
# embedded tokens, before passing it to the transformer
embedded = tied_embedding_softmax(tokens, embed=True)
# the activations after passing it from the transformer
# for some odd reason, TPUs don't play well with specifying the arguments of the Encoder() function
# so you have to leave them at their defaults
transformed = transformer.Encoder()(embedded, training=False)
# pass the activations from our tiedsoftmax class
# this time with embed=False denoting that we are doing the softmax operation
# and not a lookup
logits = tied_embedding_softmax(transformed, embed=False)
# finally, define the Keras model with inputs as tokens and outputs as the logits we just computed
model = tf.keras.Model(inputs=tokens, outputs=logits)
# the optimizer is not used since this code only supports inference
# however, to compile the model, we still define it
optimizer = tf.contrib.tpu.CrossShardOptimizer(
tf.contrib.estimator.clip_gradients_by_norm(
tf.train.AdagradOptimizer(learning_rate=1e-2), 0.25)
)
# compile the model with the optimizer and loss
model.compile(optimizer=optimizer, loss=loss)
print(model.summary())
# IMPORTANT
# this is where the saved model is presented to the code
# the model directory should have the model checkpoint and
# a checkpoint file
run_config = tf.contrib.tpu.RunConfig(
model_dir=model_dir)
# this converts the Keras model to a TensorFlow estimator
# this step is critical
# remember to patch the TF 1.14 file before running the code, else you're going to see errors here
estimator_model = tf.keras.estimator.model_to_estimator(keras_model=model, config=run_config)
# we now create a serving function from this estimator
# this enables us to load the model once and easily query it multiple times
def serving_input_fn():
inputs = {'input_1': tf.placeholder(tf.int32, [1, self.seq_length])}
return tf.estimator.export.ServingInputReceiver(inputs, inputs)
self.predict_fn = tf.contrib.predictor.from_estimator(estimator_model, serving_input_fn)
# almost there, we now take the user prompt and tokenize with BPE
# load BPE codes
self.bpe = fastBPE.fastBPE(code_file, vocab_file)
self.temperature=temperature
self.nucleusprob = nucleus_prob
self.penalty = 1.1
self.topk=topk
def configure_verb_probs(self, probabilities, options):
# Make sure only a possible verb is chosen.
for word in get_possible_verbs():
probabilities[self.word2idx[word]] += 100
# Disallow used verbs
if "used_verbs" in options:
for verb in options["used_verbs"]:
if verb in self.word2idx:
probabilities[self.word2idx[verb]] = -1e8
return probabilities
def prompt_replace(self, prompt):
# print("\n\nBEFORE PROMPT_REPLACE:")
# print(repr(prompt))
if prompt[-1] != " ":
prompt = prompt + " "
prompt = second_to_first_person(prompt)
prompt = self.control_code + prompt
# print("\n\nAFTER PROMPT_REPLACE")
# print(repr(prompt))
return prompt
def result_replace(self, result):
# print("\n\nBEFORE RESULT_REPLACE:")
# print(repr(result))
result = cut_trailing_sentence(result)
first_letter_capitalized = result[0].isupper()
result = result.replace('."', '".')
result = result.replace("#", "")
result = result.replace("*", "")
result = first_to_second_person(result)
result = remove_profanity(result)
if not first_letter_capitalized:
result = result[0].lower() + result[1:]
while("\n \n \n " in result):
result = result.replace("\n \n \n ", "\n \n ")
# print("\n\nAFTER RESULT_REPLACE:")
# print(repr(result))
return result
def generate_next_token(self, token, tokens_generated, options, num_new_lines, token_num, first_token=False, forbid_newline=False):
# get the logits from the prediction function
# the logic here is a bit convoluted because we are allowing generation past 512 tokens
# this is done by sliding the window over (past 512 tokens) and continuing prediction
# I'm sure this can be simplified (TODO)
if token <= self.seq_length:
prompt_logits = self.predict_fn({'input_1': tokens_generated[:, :self.seq_length]})[
'tied_embedding_softmax'].squeeze() / (self.temperature if self.temperature > 0 else 1.)
_token = token if token < self.seq_length else -1
else:
_token = -1
end = token + 1
start = token - self.seq_length + 2
prompt_logits = \
self.predict_fn({'input_1': np.hstack((tokens_generated[:, 0:1], tokens_generated[:, start:end]))})[
'tied_embedding_softmax'].squeeze() / (self.temperature if self.temperature > 0 else 1.)
# if penalty (for repetition) is non-zero,
# discount the logits from already generated tokens
if self.penalty > 0:
penalized_so_far = set()
for _ in range(token + 1):
generated_token = tokens_generated[0][_]
penalized_so_far.add(generated_token)
prompt_logits[_token][generated_token] /= self.penalty
# disallow some tokens
forbidden_tokens = ['<unk>', 'Sco@@', "&amp@@", "1]@@", "2]@@", "3]@@", "4]@@", "https://www.@@", "[@@", ":@@",
"Edit", "&@@", "2:","1:", ":", "Edit@@", "EDI@@", "EDIT@@", "edit", "TL@@", "tl@@", ";@@",
'**', "http://@@", "Redd@@", "UP@@", "mom", "Up@@", "Me:", "Update", "mom@@", "Part",
"http://www.@@", "edit@@", "*@@", "Writing", "Text@@", "\\@@", "<br>@@", "<div", "|@@", '...']
for forbidden_token in forbidden_tokens:
prompt_logits[_token][self.word2idx[forbidden_token]] = -1e8
if forbid_newline:
prompt_logits[_token][self.word2idx['\n']] = -1e8
else:
prompt_logits[_token][self.word2idx['\n']] *= 1.0
# Set whitelist
if "word_whitelist" in options and token_num in options["word_whitelist"].keys():
for word in options["word_whitelist"][token_num]:
prompt_logits[_token][self.word2idx[word]] += 100
# Set blacklist, overwrites whitelist
if "word_blacklist" in options and token_num in options["word_blacklist"].keys():
for word in options["word_blacklist"][token_num]:
prompt_logits[_token][self.word2idx[word]] = -1e8
# compute probabilities from logits
prompt_probs = np.exp(prompt_logits[_token])
prompt_probs = prompt_probs / sum(prompt_probs)
pruned_list = np.argsort(prompt_probs)[::-1]
# if you are using nucleus prob, then compute the nucleus probability size
if self.nucleusprob > 0.:
minimum_topk = 1
nucleus = max(np.where(np.cumsum(np.sort(prompt_probs)[::-1]) > self.nucleusprob)[0][0], minimum_topk)
elif self.topk > 0:
nucleus = self.topk
else:
nucleus = len(pruned_list)
pruned_list = pruned_list[:nucleus]
# if temperature is 0
# just pick the first (most probable) token
if self.temperature == 0:
idx = pruned_list[0]
else:
# else,
# sample from the pruned_list with the logits
chosen_idx = int(
tf.random.categorical(np.expand_dims(prompt_logits[_token][pruned_list], 0), num_samples=1).numpy())
idx = pruned_list[chosen_idx]
return idx
def generate(self, prompt, options=None):
prompt = self.prompt_replace(prompt)
debug_print = True
if debug_print:
print("\n\n*****DEBUG*****")
print("Prompt is:")
print(prompt + "\n\n")
if options is None: options = dict()
if "used_verbs" not in options:
options["used_verbs"] = set()
first_token = True
# tokenize provided prompt
split_prompt = self.bpe.apply([prompt])[0].split()
text = [self.word2idx[i] for i in split_prompt]
total_text_len = len(text) + self.generate_num
# pad with 0s and create a mini-batch of 2 (arbitrary, for ease of code)
padded_text = text + [0] * (total_text_len - len(text))
tokens_generated = np.tile(padded_text, (1, 1))
result = ""
token_num = 0
num_new_lines = 0
for token in range(len(text) - 1, total_text_len - 1):
idx = self.generate_next_token(token, tokens_generated, options, num_new_lines, token_num, first_token=first_token, forbid_newline=False)
is_nothing = len(cut_trailing_sentence(result)) == 0 or len(cut_trailing_quotes(result)) == 0
if self.idx2word[idx] == '\n' and token_num > 7 and not is_nothing:
return self.result_replace(result)
elif self.idx2word[idx] == '\n':
idx = self.generate_next_token(token, tokens_generated, options, num_new_lines, token_num,
first_token=first_token, forbid_newline=True)
# assign the token for generation
tokens_generated[0][token + 1] = idx
if debug_print:
print(repr(self.idx2word[idx]), end="_")
tokens_generated_so_far = ' '.join([self.idx2word[c] for c in tokens_generated[0][len(text):].squeeze()[:token + 2]])
tokens_generated_so_far = re.sub('(@@ )', '', string=tokens_generated_so_far)
tokens_generated_so_far = re.sub('(@@ ?$)', '', string=tokens_generated_so_far)
result = tokens_generated_so_far
token_num += 1
if debug_print:
print("\n****END DEBUG*****\n")
return self.result_replace(result)