Spaces:
Runtime error
Runtime error
Delete streamlit_app.py
Browse files- streamlit_app.py +0 -255
streamlit_app.py
DELETED
@@ -1,255 +0,0 @@
|
|
1 |
-
import streamlit as st
|
2 |
-
import numpy as np
|
3 |
-
import pandas as pd
|
4 |
-
|
5 |
-
# modeling
|
6 |
-
import torch
|
7 |
-
from torch.utils.data import Dataset, DataLoader
|
8 |
-
import pytorch_lightning as pl
|
9 |
-
from pytorch_lightning import Trainer, seed_everything
|
10 |
-
from pytorch_lightning.callbacks import ModelCheckpoint
|
11 |
-
from pytorch_lightning.callbacks.early_stopping import EarlyStopping
|
12 |
-
from transformers import (
|
13 |
-
T5ForConditionalGeneration,
|
14 |
-
T5TokenizerFast as T5Tokenizer,
|
15 |
-
)
|
16 |
-
from transformers.optimization import Adafactor
|
17 |
-
|
18 |
-
# aesthetics
|
19 |
-
from IPython.display import Markdown, display, clear_output
|
20 |
-
import re
|
21 |
-
import warnings
|
22 |
-
warnings.filterwarnings(
|
23 |
-
"ignore", ".*Trying to infer the `batch_size` from an ambiguous collection.*"
|
24 |
-
)
|
25 |
-
seed_everything(25429)
|
26 |
-
|
27 |
-
# scoring
|
28 |
-
import spacy
|
29 |
-
from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
|
30 |
-
|
31 |
-
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
32 |
-
print(device)
|
33 |
-
|
34 |
-
# loading the model
|
35 |
-
hug = 't5-small'
|
36 |
-
t5tokenizer = T5Tokenizer.from_pretrained(hug)
|
37 |
-
t5model = T5ForConditionalGeneration.from_pretrained(hug, return_dict=True)
|
38 |
-
|
39 |
-
# defining tokens
|
40 |
-
SEP_TOKEN = '<sep>'
|
41 |
-
MASK_TOKEN = '[MASK]'
|
42 |
-
MASKING_CHANCE = 0.1
|
43 |
-
|
44 |
-
class DataEncodings(Dataset):
|
45 |
-
'''
|
46 |
-
tokenizes, pads, and adds special tokens
|
47 |
-
'''
|
48 |
-
def __init__(
|
49 |
-
self,
|
50 |
-
data: pd.DataFrame,
|
51 |
-
tokenizer,
|
52 |
-
source_max_token_len: int,
|
53 |
-
target_max_token_len: int
|
54 |
-
):
|
55 |
-
self.tokenizer = t5tokenizer
|
56 |
-
self.data = data
|
57 |
-
self.source_max_token_len = source_max_token_len
|
58 |
-
self.target_max_token_len = target_max_token_len
|
59 |
-
|
60 |
-
def __len__(self):
|
61 |
-
return len(self.data)
|
62 |
-
|
63 |
-
def __getitem__(self, index:int):
|
64 |
-
data_row = self.data.iloc[index]
|
65 |
-
# adds a random mask for answer-agnostic qg
|
66 |
-
if np.random.rand() > MASKING_CHANCE:
|
67 |
-
answer = data_row['answer']
|
68 |
-
else:
|
69 |
-
answer = MASK_TOKEN
|
70 |
-
|
71 |
-
source_encoding = t5tokenizer(
|
72 |
-
f"{answer} {SEP_TOKEN} {data_row['context']}",
|
73 |
-
max_length= self.source_max_token_len,
|
74 |
-
padding='max_length',
|
75 |
-
truncation= True,
|
76 |
-
return_attention_mask=True,
|
77 |
-
return_tensors='pt'
|
78 |
-
)
|
79 |
-
|
80 |
-
target_encoding = t5tokenizer(
|
81 |
-
f"{data_row['answer']} {SEP_TOKEN} {data_row['question']}",
|
82 |
-
max_length=self.target_max_token_len,
|
83 |
-
padding='max_length',
|
84 |
-
truncation = True,
|
85 |
-
return_attention_mask=True,
|
86 |
-
return_tensors='pt'
|
87 |
-
)
|
88 |
-
|
89 |
-
labels = target_encoding['input_ids']
|
90 |
-
labels[labels == 0] = -100 # masked
|
91 |
-
|
92 |
-
encodings = dict(
|
93 |
-
answer = data_row['answer'],
|
94 |
-
context = data_row['context'],
|
95 |
-
question = data_row['question'],
|
96 |
-
input_ids = source_encoding['input_ids'].flatten(),
|
97 |
-
attention_mask = source_encoding['attention_mask'].flatten(),
|
98 |
-
labels=labels.flatten()
|
99 |
-
)
|
100 |
-
|
101 |
-
return encodings
|
102 |
-
|
103 |
-
class DataModule(pl.LightningDataModule):
|
104 |
-
|
105 |
-
def __init__(
|
106 |
-
self,
|
107 |
-
train: pd.DataFrame,
|
108 |
-
val: pd.DataFrame,
|
109 |
-
tokenizer,
|
110 |
-
batch_size,
|
111 |
-
source_max_token_len: int,
|
112 |
-
target_max_token_len: int
|
113 |
-
):
|
114 |
-
super().__init__()
|
115 |
-
self.batch_size = batch_size
|
116 |
-
self.train = train
|
117 |
-
self.val = val
|
118 |
-
self.tokenizer = t5tokenizer
|
119 |
-
self.source_max_token_len = source_max_token_len
|
120 |
-
self.target_max_token_len = target_max_token_len
|
121 |
-
|
122 |
-
def setup(self):
|
123 |
-
self.train_dataset = DataEncodings(self.train, self.tokenizer, self.source_max_token_len, self.target_max_token_len)
|
124 |
-
self.val_dataset = DataEncodings(self.val, self.tokenizer, self.source_max_token_len, self.target_max_token_len)
|
125 |
-
|
126 |
-
def train_dataloader(self):
|
127 |
-
return DataLoader(self.train_dataset, batch_size=batch_size, shuffle=True, num_workers=0)
|
128 |
-
|
129 |
-
def val_dataloader(self):
|
130 |
-
return DataLoader(self.val_dataset, batch_size=batch_size, num_workers=0)
|
131 |
-
|
132 |
-
# hyperparameters
|
133 |
-
num_epochs = 16
|
134 |
-
batch_size = 32
|
135 |
-
learning_rate = 0.001
|
136 |
-
|
137 |
-
# model
|
138 |
-
class T5Model(pl.LightningModule):
|
139 |
-
def __init__(self):
|
140 |
-
super().__init__()
|
141 |
-
self.model = t5model
|
142 |
-
self.model.resize_token_embeddings(len(t5tokenizer)) # resizing after adding new tokens to the tokenizer
|
143 |
-
|
144 |
-
# feed forward pass
|
145 |
-
def forward(self, input_ids, attention_mask, labels=None):
|
146 |
-
output = self.model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)
|
147 |
-
return output.loss, output.logits
|
148 |
-
|
149 |
-
# train model and compute loss
|
150 |
-
def training_step(self, batch, batch_idx):
|
151 |
-
input_ids = batch['input_ids']
|
152 |
-
attention_mask = batch['attention_mask']
|
153 |
-
labels = batch['labels']
|
154 |
-
loss, output = self(input_ids, attention_mask, labels)
|
155 |
-
self.log('train_loss', loss, prog_bar=True, logger=True, batch_size=batch_size)
|
156 |
-
return loss
|
157 |
-
|
158 |
-
# gets model predictions, returns loss
|
159 |
-
def validation_step(self, batch, batch_idx):
|
160 |
-
input_ids = batch['input_ids']
|
161 |
-
attention_mask = batch['attention_mask']
|
162 |
-
labels = batch['labels']
|
163 |
-
loss, output = self(input_ids, attention_mask, labels)
|
164 |
-
self.log('val_loss', loss, prog_bar=True, logger=True, batch_size=batch_size)
|
165 |
-
return {'val loss': loss}
|
166 |
-
|
167 |
-
# def validation_epoch_end(self, outputs):
|
168 |
-
# # outputs = list of dictionaries to print loss
|
169 |
-
# avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
|
170 |
-
# tensorboard_logs = {'avg_val_loss': avg_loss}
|
171 |
-
# return {'val_loss': avg_loss, 'log': tensorboard_logs}
|
172 |
-
|
173 |
-
def configure_optimizers(self):
|
174 |
-
return Adafactor(model.parameters(), scale_parameter=False, relative_step=False, lr=learning_rate)
|
175 |
-
|
176 |
-
def generate(model: T5Model, answer:str, context:str, beams, length, temper) -> str:
|
177 |
-
source_encoding = t5tokenizer(
|
178 |
-
f"{answer} {SEP_TOKEN} {context}",
|
179 |
-
max_length=512,
|
180 |
-
padding='max_length',
|
181 |
-
truncation=True,
|
182 |
-
return_attention_mask=True,
|
183 |
-
add_special_tokens=True,
|
184 |
-
return_tensors='pt'
|
185 |
-
)
|
186 |
-
|
187 |
-
generated_ids=model.model.generate(
|
188 |
-
input_ids=source_encoding['input_ids'],
|
189 |
-
attention_mask=source_encoding['attention_mask'],
|
190 |
-
num_beams=beams,
|
191 |
-
max_length=length,
|
192 |
-
repetition_penalty=2.5,
|
193 |
-
length_penalty=0.8,
|
194 |
-
temperature=temper,
|
195 |
-
early_stopping=True,
|
196 |
-
use_cache=True
|
197 |
-
)
|
198 |
-
|
199 |
-
preds = {
|
200 |
-
t5tokenizer.decode(generated_id, skip_special_tokens=False, clean_up_tokenization_spaces=True)
|
201 |
-
for generated_id in generated_ids
|
202 |
-
}
|
203 |
-
|
204 |
-
return ''.join(preds)
|
205 |
-
|
206 |
-
def show_result(generated:str, answer:str, context:str, original_question:str=''):
|
207 |
-
|
208 |
-
regex = r"(?<=>)(.*?)(?=<)"
|
209 |
-
matches = re.findall(regex, generated)
|
210 |
-
matches[1] = matches[1][5:]
|
211 |
-
final = {cat: match.strip() for cat, match in zip(['Answer', 'Question'], matches)}
|
212 |
-
st.title('Context')
|
213 |
-
st.write(context)
|
214 |
-
st.title('Answer')
|
215 |
-
st.write(answer)
|
216 |
-
st.title('Generated')
|
217 |
-
st.write(final)
|
218 |
-
# if original_question:
|
219 |
-
# printBold('Original Question')
|
220 |
-
# print(original_question)
|
221 |
-
# gen = nlp(matches[1])
|
222 |
-
# ori = nlp(original_question)
|
223 |
-
# bleu_score = sentence_bleu(matches[1], original_question, smoothing_function=SmoothingFunction().method5)
|
224 |
-
# cs_score = ori.similarity(gen)
|
225 |
-
# printBold('Scores')
|
226 |
-
# print(f"BLEU: {bleu_score}")
|
227 |
-
# print(f'Cosine Similarity: {cs_score}')
|
228 |
-
# return bleu_score, cs_score
|
229 |
-
|
230 |
-
# streamlit app
|
231 |
-
st.title('Question Generation From Text')
|
232 |
-
|
233 |
-
with st.form('my_form'):
|
234 |
-
context = st.text_input('Enter a context passage for question generation:', 'The capital of France is Paris.')
|
235 |
-
answer = st.text_input('Give a correct answer, or [MASK] for unsupervised generation:', 'Paris')
|
236 |
-
# question = st.text_input('Question', 'What is the capital of France?')
|
237 |
-
# original_question = st.text_input('Original Question', 'What is the capital of France?')
|
238 |
-
beams = st.sidebar.slider('Beams', min_value=1, max_value=20)
|
239 |
-
length = st.sidebar.slider('Maximum length of generated question', min_value=50, max_value=200)
|
240 |
-
temper = st.sidebar.slider("Temperature", value = 1.0, min_value = 0.0, max_value=1.0, step=0.05)
|
241 |
-
submitted = st.form_submit_button('Generate')
|
242 |
-
|
243 |
-
with st.spinner('Loading Model...'):
|
244 |
-
model = T5Model
|
245 |
-
best_model_dir = 't5-chkpt-v2.ckpt'
|
246 |
-
best_model = model.load_from_checkpoint(best_model_dir)
|
247 |
-
# best_model = model.load_from_checkpoint(callback.best_model_path)
|
248 |
-
best_model.freeze()
|
249 |
-
|
250 |
-
with st.spinner('Generating...'):
|
251 |
-
if submitted:
|
252 |
-
|
253 |
-
generated = generate(best_model, answer, context, beams, length, temper)
|
254 |
-
show_result(generated, answer, context)
|
255 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|