Spaces:
Running
Running
Commit
·
88fdd1b
1
Parent(s):
2f235b9
add app
Browse files
app.py
ADDED
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from datasets import load_dataset
|
3 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
4 |
+
from time import time
|
5 |
+
import torch
|
6 |
+
|
7 |
+
@st.cache(
|
8 |
+
allow_output_mutation=True,
|
9 |
+
hash_funcs={
|
10 |
+
AutoTokenizer: lambda x: None,
|
11 |
+
AutoModelForSeq2SeqLM: lambda x: None,
|
12 |
+
},
|
13 |
+
suppress_st_warning=True
|
14 |
+
)
|
15 |
+
def load_models():
|
16 |
+
st_time = time()
|
17 |
+
tokenizer = AutoTokenizer.from_pretrained("Babelscape/rebel-large")
|
18 |
+
print("+++++ loading Model", time() - st_time)
|
19 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("Babelscape/rebel-large")
|
20 |
+
if torch.cuda.is_available():
|
21 |
+
_ = model.to("cuda:0") # comment if no GPU available
|
22 |
+
_ = model.eval()
|
23 |
+
print("+++++ loaded model", time() - st_time)
|
24 |
+
dataset = load_dataset('Babelscape/rebel-dataset', split="train[:1%]")
|
25 |
+
return (tokenizer, model, dataset)
|
26 |
+
|
27 |
+
def extract_triplets(text):
|
28 |
+
triplets = []
|
29 |
+
relation = ''
|
30 |
+
for token in text.split():
|
31 |
+
if token == "<triplet>":
|
32 |
+
current = 't'
|
33 |
+
if relation != '':
|
34 |
+
triplets.append((subject, relation, object_))
|
35 |
+
relation = ''
|
36 |
+
subject = ''
|
37 |
+
elif token == "<subj>":
|
38 |
+
current = 's'
|
39 |
+
if relation != '':
|
40 |
+
triplets.append((subject, relation, object_))
|
41 |
+
object_ = ''
|
42 |
+
elif token == "<obj>":
|
43 |
+
current = 'o'
|
44 |
+
relation = ''
|
45 |
+
else:
|
46 |
+
if current == 't':
|
47 |
+
subject += ' ' + token
|
48 |
+
elif current == 's':
|
49 |
+
object_ += ' ' + token
|
50 |
+
elif current == 'o':
|
51 |
+
relation += ' ' + token
|
52 |
+
triplets.append((subject, relation, object_))
|
53 |
+
return triplets
|
54 |
+
|
55 |
+
|
56 |
+
tokenizer, model, dataset = load_models()
|
57 |
+
|
58 |
+
agree = st.checkbox('Free input', False)
|
59 |
+
if agree:
|
60 |
+
text = st.text_input('Input text', 'Punta Cana is a resort town in the municipality of Higüey, in La Altagracia Province, the easternmost province of the Dominican Republic.')
|
61 |
+
print(text)
|
62 |
+
else:
|
63 |
+
dataset_example = st.slider('dataset id', 0, 1000, 0)
|
64 |
+
text = dataset[dataset_example]['context']
|
65 |
+
length_penalty = st.slider('length_penalty', 0, 10, 0)
|
66 |
+
num_beams = st.slider('num_beams', 1, 20, 3)
|
67 |
+
num_return_sequences = st.slider('num_return_sequences', 1, num_beams, 2)
|
68 |
+
|
69 |
+
gen_kwargs = {
|
70 |
+
"max_length": 256,
|
71 |
+
"length_penalty": length_penalty,
|
72 |
+
"num_beams": num_beams,
|
73 |
+
"num_return_sequences": num_return_sequences,
|
74 |
+
}
|
75 |
+
|
76 |
+
model_inputs = tokenizer(text, max_length=256, padding=True, truncation=True, return_tensors = 'pt')
|
77 |
+
generated_tokens = model.generate(
|
78 |
+
model_inputs["input_ids"].to(model.device),
|
79 |
+
attention_mask=model_inputs["attention_mask"].to(model.device),
|
80 |
+
**gen_kwargs,
|
81 |
+
)
|
82 |
+
|
83 |
+
decoded_preds = tokenizer.batch_decode(generated_tokens, skip_special_tokens=False)
|
84 |
+
st.title('Input text')
|
85 |
+
|
86 |
+
st.write(text)
|
87 |
+
|
88 |
+
if not agree:
|
89 |
+
st.title('Silver output')
|
90 |
+
st.write(dataset[dataset_example]['triplets'])
|
91 |
+
st.write(extract_triplets(dataset[dataset_example]['triplets']))
|
92 |
+
|
93 |
+
st.title('Prediction text')
|
94 |
+
decoded_preds = [text.replace('<s>', '').replace('</s>', '').replace('<pad>', '') for text in decoded_preds]
|
95 |
+
st.write(decoded_preds)
|
96 |
+
|
97 |
+
for idx, sentence in enumerate(decoded_preds):
|
98 |
+
st.title(f'Prediction triplets sentence {idx}')
|
99 |
+
st.write(extract_triplets(sentence))
|