-
Notifications
You must be signed in to change notification settings - Fork 26
/
interact_generator_only.py
303 lines (234 loc) · 9.9 KB
/
interact_generator_only.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
import logging
from TTS.text2speech import tts_class
from multiprocessing import Process
import faiss
import time
import sqlite3
import csv
import random
import copy
import tensorflow_hub as hub
import tensorflow_text
import math
import numpy as np
import pickle
from Retriever.Retrieve import retrieve
import Utils.functions as utils
from ReRanker.rerank import rank_and_choose
from Generator.generator import generate as DialoGPT_Generate
from Classifier.model.dialog_acts import Encoder as Classifier
from Sentence_Encoder.meta_response_encoder_fast import encode as response_encode
from Sentence_Encoder.meta_query_encoder_fast import encode as query_encode
import Sentence_Encoder.encoder_client as encoder_client
import tensorflow as tf
from transformers import GPT2Tokenizer, GPT2LMHeadModel, GPT2Config
import torch.nn.functional as F
import torch.nn as nn
import torch as T
import os
import sys
import argparse
logging.getLogger("tensorflow").setLevel(logging.CRITICAL)
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
logging.basicConfig(level=logging.CRITICAL)
parser = argparse.ArgumentParser(description="Chatbot")
parser.add_argument('--voice', dest='voice', action='store_true')
parser.add_argument('--no-voice', dest='voice', action='store_false')
parser.set_defaults(voice=True)
flags = parser.parse_args()
device = "cuda"
with open("Retriever/Faiss_index/thread_idx.pkl", 'rb') as fp:
idx = pickle.load(fp)
index = faiss.read_index('Retriever/Faiss_index/large.index')
# LOAD DATABASE
conn = sqlite3.connect('Retriever/Database/reddit.db')
c = conn.cursor()
# LOAD SCRIPTS
with open('Scripted/Processed_Scripts/Bot_Profile.pkl', 'rb') as fp:
bot_profile = pickle.load(fp)
bot_queries = [k for k, v in bot_profile.items()]
with open('Scripted/Processed_Scripts/Chatterbot.pkl', 'rb') as fp:
chatterbot = pickle.load(fp)
chatterbot_queries = [k for k, v in chatterbot.items()]
# LOAD SCRIPT EMBEDDINGS
with open('Scripted/Processed_Scripts/embedded_bot_queries.pkl', 'rb') as fp:
bot_queries_embd = pickle.load(fp)
with open('Scripted/Processed_Scripts/embedded_chatterbot_queries.pkl', 'rb') as fp:
chatterbot_queries_embd = pickle.load(fp)
# Load TTS model
with T.no_grad():
text2speech = tts_class()
# Load Dialog Acts Classifer
with open("Classifier/data/processed_data.pkl", "rb") as fp:
data = pickle.load(fp)
labels2idx = data["labels2idx"]
idx2labels = {v: k for k, v in labels2idx.items()}
with T.no_grad():
dialog_act_classifier = Classifier(
D=bot_queries_embd.shape[-1], classes_num=len(labels2idx)).cuda()
checkpoint = T.load("Classifier/Model_Backup/model.pt")
dialog_act_classifier.load_state_dict(checkpoint['model_state_dict'])
dialog_act_classifier = dialog_act_classifier.eval()
# LOAD DialoGPT Generator
with T.no_grad():
tokenizer = GPT2Tokenizer.from_pretrained('Generator/DialoGPT/Configs/')
weights = T.load('Generator/DialoGPT/Parameters/medium_ft.pkl')
weights_reverse = T.load('Generator/DialoGPT/Parameters/small_reverse.pkl')
cfg = GPT2Config.from_json_file('Generator/DialoGPT/Configs/config.json')
model = GPT2LMHeadModel(cfg)
model_reverse = GPT2LMHeadModel(cfg)
# fix misused key value
weights["lm_head.weight"] = weights["lm_head.decoder.weight"]
weights.pop("lm_head.decoder.weight", None)
weights_reverse["lm_head.weight"] = weights_reverse["lm_head.decoder.weight"]
weights_reverse.pop("lm_head.decoder.weight", None)
model.load_state_dict(weights)
model.to('cuda')
model.eval()
model_reverse.load_state_dict(weights_reverse)
model_reverse.to('cuda')
model_reverse.eval()
with tf.device("/cpu:0"):
# Hub Models
ConvRT_model = encoder_client.EncoderClient(
"Sentence_Encoder/Embeddings/ConvRT", use_extra_context=True)
USE_QA_model = hub.load("Sentence_Encoder/Embeddings/USE_QA/")
# %%
command_codes = ["<PASS>", "<JOKE>", "<GENERATE>",
"<INITIATE>", "<TIL>", "<STORY>", "<SHOWER>", "<STOP>"]
code_map = {"<INITIATE>": ["Scripted/Random_Reddit_Data/nostupidq.csv",
"Scripted/Random_Reddit_Data/jokesq.csv",
"Scripted/Random_Reddit_Data/showerthoughtsq.csv",
"Scripted/Random_Reddit_Data/tilq.csv"],
"<TIL>": ["Scripted/Random_Reddit_Data/tilq.csv"],
"<SHOWER>": ["Scripted/Random_Reddit_Data/showerthoughtsq.csv"],
"<STORY>": ["Scripted/Random_Reddit_Data/writingpromptsa.csv"],
"<JOKE>": ["Scripted/Random_Reddit_Data/jokesq.csv"]}
def random_response(candidates, conversation_history, p=None):
loop = 5
if p is None:
response = random.choice(candidates)
else:
response = np.random.choice(candidates, p=p)
i = 0
while response in conversation_history:
if p is None:
response = random.choice(candidates)
else:
response = np.random.choice(candidates, p=p)
i += 1
if i > loop:
break
return response
# %%
def load_random_reddit(directory, conversation_history):
candidates = []
with open(directory, newline='') as csvfile:
csv_reader = csv.DictReader(csvfile)
for i, row in enumerate(csv_reader):
if 'writing' in directory:
parent_id = str(row['parent_id'])[3:]
thread_id = str(row['link_id'])[3:]
if parent_id == thread_id:
candidate = str(row["body"])
else:
candidate = str(row["title"])
if 'joke' in directory:
candidate += ".... "+str(row['selftext'])
candidates.append(candidate)
return random_response(candidates, conversation_history)
# extract top candidates (queries or responses)
def top_candidates(candidates, scores, top=1):
sorted_score_idx = np.flip(np.argsort(scores), axis=-1)
candidates = [candidates[i] for i in sorted_score_idx.tolist()]
scores = [scores[i] for i in sorted_score_idx.tolist()]
return candidates[0:top], scores[0:top], sorted_score_idx.tolist()
# %%
def generate(texts, past):
candidates, _ = DialoGPT_Generate(texts, model, tokenizer)
return candidates, past
# START DOING STUFF
conversation_history = []
past = None
stop_flag = 0
print("\n")
while True:
utterance = input("Say Something: ") # ,hello how are ya today"
utils.delay_print("\nThinking......")
candidates = []
temp_candidates = []
temp_scores = []
if not conversation_history:
query_context = []
response_context = [""]
else:
if len(conversation_history) > 5:
truncated_history = copy.deepcopy(conversation_history[-5:])
else:
truncated_history = copy.deepcopy(conversation_history)
response_context = [conversation_history[-1]]
# ConveRT needs reversed Context, not sure about USE QA but assuming it's not reverse
query_context = [stuff for stuff in truncated_history]
query_encoding = query_encode([utterance], USE_QA_model, ConvRT_model, [query_context])
if conversation_history:
if len(conversation_history) > 5:
truncated_history = conversation_history[-5:]
else:
truncated_history = conversation_history
generated_responses, past = generate(truncated_history+[utterance], past)
else:
generated_responses, past = generate([utterance], past)
with T.no_grad():
logits = dialog_act_classifier(T.tensor(query_encoding).to(device))
_, sorted_idx = T.sort(logits, dim=-1, descending=True)
sorted_idx = sorted_idx.squeeze(0)
sorted_idx = sorted_idx[0:2].cpu().tolist()
labels = [idx2labels[i] for i in sorted_idx]
bias = None
if bias is not None:
biases = [0.0 for _ in candidates]
for _ in generated_responses:
biases.append(0.0)
for _ in retrieved_candidates:
biases.append(bias)
biases = np.asarray(biases, np.float32)
else:
biases = None
candidates = generated_responses
response, _ = rank_and_choose(USE_QA_model, ConvRT_model,
tokenizer,
model_reverse,
utterance,
query_encoding,
candidates,
response_context,
conversation_history,
bias=biases)
print("\n")
if len(str(response).split(" ")) <= 100:
if flags.voice:
entry = utils.simple_preprocess(str(response).lower(),
for_speech=True,
return_tokenized=True)
entry = " ".join(entry)
wavefiles = text2speech.process(entry)
def f1():
utils.delay_print("Bot: "+response)
def f2():
text2speech.play(wavefiles)
p1 = Process(target=f1)
p2 = Process(target=f2)
p1.start()
p2.start()
p1.join()
p2.join()
else:
utils.delay_print("Bot: "+response)
else:
utils.delay_print("Bot: "+response, t=0.01)
print("\n")
conversation_history.append(utterance)
conversation_history.append(response)
if stop_flag == 1:
break
# break