-
Notifications
You must be signed in to change notification settings - Fork 0
/
demo.py
191 lines (146 loc) · 4.99 KB
/
demo.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
import json
import os
import re
import yaml
from pyserini.search import SimpleSearcher
import torch
from transformers import T5ForConditionalGeneration, T5TokenizerFast
import streamlit as st
from unidecode import unidecode
from model import get_model
@st.cache(allow_output_mutation=True, suppress_st_warning=True)
def init():
# Configuration
with open("config/demo.yaml", "r") as f:
conf = yaml.safe_load(f)
# Device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Setup question rewriting model
tokenizer_t5 = T5TokenizerFast.from_pretrained(conf["rewrite_model_name"])
t5 = T5ForConditionalGeneration.from_pretrained(conf["rewrite_model_name"])
t5.to(device)
t5.eval()
# Setup retrieval model
searcher = SimpleSearcher(conf["passages"])
searcher.set_bm25(0.82, 0.68)
# Setup answer generation model
model = get_model(conf["model_name"]).load_from_checkpoint(conf["from_checkpoint"])
tokenizer_pegasus = model.tokenizer
pegasus = model.model
pegasus.to(device)
pegasus.eval()
torch.no_grad()
return {
"conf": conf,
"device": device,
"tokenizer_t5": tokenizer_t5,
"t5": t5,
"searcher": searcher,
"tokenizer_pegasus": tokenizer_pegasus,
"pegasus": pegasus,
}
@st.cache(allow_output_mutation=True)
def Text():
return ["Q: "]
def compute_answer(model, history):
# REWRITE
while True:
# Tokenize
rewrite_input = " ||| ".join(history[-model["conf"]["rewrite_max_history"] :])
rewrite_input_ids = model["tokenizer_t5"].encode(
rewrite_input,
truncation=False,
return_tensors="pt",
)
if len(rewrite_input_ids) <= model["conf"]["rewrite_max_input_length"]:
break
else:
del history[: -model["conf"]["rewrite_max_history"] + 1]
rewrite_input_ids = rewrite_input_ids.to(model["device"])
output = model["t5"].generate(
rewrite_input_ids,
max_length=model["conf"]["rewrite_max_output_length"],
do_sample=True,
)
model_rewrite = model["tokenizer_t5"].batch_decode(
output,
skip_special_tokens=True,
clean_up_tokenization_spaces=True,
)[0]
history[-1] = model_rewrite
# RETRIEVE
query = "\n".join(history[-model["conf"]["max_history"] :])
hits = model["searcher"].search(query, k=model["conf"]["max_candidates"])[
: model["conf"]["max_candidates"]
]
model_passages = {hit.docid: hit.score for hit in hits}
passages = [
json.loads(model["searcher"].doc(docid).raw())["contents"]
for docid in model_passages
]
# GENERATE
context = query
generate_input = "\n\n".join([context] + passages)
generate_input = generate_input.encode("ascii", "ignore").decode("utf8")
generate_input = unidecode(generate_input)
generate_input = generate_input.replace("\n", "<n>")
generate_input_ids = (
model["tokenizer_pegasus"]
.encode(
generate_input,
truncation=True,
max_length=model["conf"]["max_input_length"],
return_tensors="pt",
)
.to(model["device"])
)
model_input = model["tokenizer_pegasus"].batch_decode(
generate_input_ids,
)[0]
model_input = model_input.replace("<n>", "\n")
text_passages = model_input[model_input.find("\n\n") + 1 :]
output = model["pegasus"].generate(
generate_input_ids,
max_length=model["conf"]["max_output_length"],
do_sample=True,
no_repeat_ngram_size=model["conf"]["no_repeat_ngram_size"],
)
model_answer = model["tokenizer_pegasus"].batch_decode(
output,
skip_special_tokens=True,
clean_up_tokenization_spaces=True,
)[0]
history.append(model_answer)
return model_rewrite, text_passages, model_answer
st.title('Ask Me "Anything"!')
st.header("Conversational Question Answering")
st.subheader("https://ama.goncaloraposo.com")
model = init()
left, right = st.columns(2)
compute = left.button("Compute")
reset = right.button("Reset")
text_cache = Text()
textbox = st.empty()
if reset:
print("reset")
text_cache[0] = "Q: "
if compute:
print("compute")
history = text_cache[0]
history = history.split("\n")
if any([sentence.startswith("R: ") for sentence in history]):
marker = "R: "
else:
marker = "Q: "
for i, sentence in enumerate(history):
if sentence.startswith("Q") and "=>" in sentence:
history[i] = "Q: " + sentence.split("=>")[-1].strip()
history = [re.sub(r"^.*:\s?", "", sentence) for sentence in history]
print(history)
# If there is a new question
if history[-1]:
rewrite, passages, answer = compute_answer(model, history)
text_cache[0] += f" => {rewrite}\nA: {answer}\nQ: "
st.caption("Retrieved Passages")
st.write(passages)
text_cache[0] = textbox.text_area("Conversation", text_cache[0], height=480)