Spaces:
Sleeping
Sleeping
Lautaro Cardarelli
commited on
Commit
·
f34392a
1
Parent(s):
471e321
add question generatiton
Browse files- app.py +88 -2
- requirements.txt +2 -1
app.py
CHANGED
@@ -1,11 +1,97 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
2 |
from transformers import BartForConditionalGeneration
|
3 |
from transformers import BartTokenizer
|
|
|
4 |
|
5 |
tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn')
|
6 |
model = BartForConditionalGeneration.from_pretrained('facebook/bart-large-cnn')
|
7 |
|
8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
def generate_summary(text):
|
10 |
inputs = tokenizer.encode("summarize: " + text, return_tensors="pt", max_length=1024, truncation=True)
|
11 |
summary_ids = model.generate(inputs, max_length=150, min_length=50, length_penalty=2.0, num_beams=4, early_stopping=True)
|
@@ -14,9 +100,9 @@ def generate_summary(text):
|
|
14 |
|
15 |
|
16 |
def process(text):
|
17 |
-
return generate_summary(text)
|
18 |
|
19 |
|
20 |
textbox = gr.Textbox(label="Pega el text aca:", placeholder="Texto...", lines=15)
|
21 |
-
demo = gr.Interface(fn=process, inputs=textbox, outputs="text")
|
22 |
demo.launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
import pandas as pd
|
3 |
+
import torch
|
4 |
+
from googletrans import Translator
|
5 |
+
from transformers import T5Tokenizer
|
6 |
+
from transformers import T5ForConditionalGeneration
|
7 |
from transformers import BartForConditionalGeneration
|
8 |
from transformers import BartTokenizer
|
9 |
+
from transformers import pipeline
|
10 |
|
11 |
tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn')
|
12 |
model = BartForConditionalGeneration.from_pretrained('facebook/bart-large-cnn')
|
13 |
|
14 |
|
15 |
+
|
16 |
+
from transformers import PreTrainedModel
|
17 |
+
from transformers import PreTrainedTokenizer
|
18 |
+
|
19 |
+
# Question launcher
|
20 |
+
class E2EQGPipeline:
|
21 |
+
def __init__(
|
22 |
+
self,
|
23 |
+
model: PreTrainedModel,
|
24 |
+
tokenizer: PreTrainedTokenizer
|
25 |
+
):
|
26 |
+
|
27 |
+
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
28 |
+
|
29 |
+
self.model = model
|
30 |
+
self.tokenizer = tokenizer
|
31 |
+
|
32 |
+
self.model_type = "t5"
|
33 |
+
|
34 |
+
self.kwargs = {
|
35 |
+
"max_length": 256,
|
36 |
+
"num_beams": 4,
|
37 |
+
"length_penalty": 1.5,
|
38 |
+
"no_repeat_ngram_size": 3,
|
39 |
+
"early_stopping": True,
|
40 |
+
}
|
41 |
+
|
42 |
+
def generate_questions(self, context: str):
|
43 |
+
inputs = self._prepare_inputs_for_e2e_qg(context)
|
44 |
+
|
45 |
+
outs = self.model.generate(
|
46 |
+
input_ids=inputs['input_ids'].to(self.device),
|
47 |
+
attention_mask=inputs['attention_mask'].to(self.device),
|
48 |
+
**self.kwargs
|
49 |
+
)
|
50 |
+
|
51 |
+
prediction = self.tokenizer.decode(outs[0], skip_special_tokens=True)
|
52 |
+
|
53 |
+
questions = prediction.split("<sep>")
|
54 |
+
questions = [question.strip() for question in questions[:-1]]
|
55 |
+
return questions
|
56 |
+
|
57 |
+
def _prepare_inputs_for_e2e_qg(self, context):
|
58 |
+
source_text = f"generate questions: {context}"
|
59 |
+
|
60 |
+
inputs = self._tokenize([source_text], padding=False)
|
61 |
+
|
62 |
+
return inputs
|
63 |
+
|
64 |
+
def _tokenize(
|
65 |
+
self,
|
66 |
+
inputs,
|
67 |
+
padding=True,
|
68 |
+
truncation=True,
|
69 |
+
add_special_tokens=True,
|
70 |
+
max_length=512
|
71 |
+
):
|
72 |
+
inputs = self.tokenizer.batch_encode_plus(
|
73 |
+
inputs,
|
74 |
+
max_length=max_length,
|
75 |
+
add_special_tokens=add_special_tokens,
|
76 |
+
truncation=truncation,
|
77 |
+
padding="max_length" if padding else False,
|
78 |
+
pad_to_max_length=padding,
|
79 |
+
return_tensors="pt"
|
80 |
+
)
|
81 |
+
|
82 |
+
return inputs
|
83 |
+
|
84 |
+
|
85 |
+
def generate_questions(text):
|
86 |
+
qg_model = T5ForConditionalGeneration.from_pretrained('valhalla/t5-base-e2e-qg')
|
87 |
+
qg_tokenizer = T5Tokenizer.from_pretrained('valhalla/t5-base-e2e-qg')
|
88 |
+
qg_final_model = E2EQGPipeline(qg_model, qg_tokenizer)
|
89 |
+
questions = qg_final_model.generate_questions(text)
|
90 |
+
translator = Translator()
|
91 |
+
translated_questions = [translator.translate(question, dest='es').text for question in questions]
|
92 |
+
return translated_questions
|
93 |
+
|
94 |
+
|
95 |
def generate_summary(text):
|
96 |
inputs = tokenizer.encode("summarize: " + text, return_tensors="pt", max_length=1024, truncation=True)
|
97 |
summary_ids = model.generate(inputs, max_length=150, min_length=50, length_penalty=2.0, num_beams=4, early_stopping=True)
|
|
|
100 |
|
101 |
|
102 |
def process(text):
|
103 |
+
return generate_summary(text), generate_questions(text)
|
104 |
|
105 |
|
106 |
textbox = gr.Textbox(label="Pega el text aca:", placeholder="Texto...", lines=15)
|
107 |
+
demo = gr.Interface(fn=process, inputs=textbox, outputs=["text", "text"])
|
108 |
demo.launch()
|
requirements.txt
CHANGED
@@ -1,3 +1,4 @@
|
|
1 |
transformers
|
2 |
torch
|
3 |
-
accelerate
|
|
|
|
1 |
transformers
|
2 |
torch
|
3 |
+
accelerate
|
4 |
+
googletrans
|