Spaces:
Runtime error
Runtime error
Añadiendo las traducciones
Browse files
app.py
CHANGED
@@ -7,15 +7,69 @@ os.system('pip install --upgrade pip')
|
|
7 |
os.system('pip install tensorflow')
|
8 |
|
9 |
from transformers import pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
docs = None
|
12 |
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
def request_pathname(files):
|
15 |
if files is None:
|
16 |
return [[]]
|
17 |
return [[file.name, file.name.split('/')[-1]] for file in files]
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
def validate_dataset(dataset):
|
21 |
global docs
|
@@ -27,19 +81,19 @@ def validate_dataset(dataset):
|
|
27 |
return "⚠️Esperando documentos..."
|
28 |
|
29 |
def do_ask(question, button, dataset):
|
30 |
-
|
31 |
global docs
|
32 |
docs_ready = dataset.iloc[-1, 0] != ""
|
33 |
if button == "✨Listo✨" and docs_ready:
|
34 |
for _, row in dataset.iterrows():
|
35 |
path = row['filepath']
|
36 |
text = Path(f'{path}').read_text()
|
|
|
37 |
question_answerer = pipeline("question-answering", model='distilbert-base-cased-distilled-squad')
|
38 |
QA_input = {
|
39 |
-
'question': question,
|
40 |
-
'context':
|
41 |
}
|
42 |
-
return question_answerer(QA_input)['answer']
|
43 |
else:
|
44 |
return ""
|
45 |
|
|
|
7 |
os.system('pip install tensorflow')
|
8 |
|
9 |
from transformers import pipeline
|
10 |
+
from transformers import MarianMTModel, MarianTokenizer
|
11 |
+
from nltk.tokenize import sent_tokenize
|
12 |
+
from nltk.tokenize import LineTokenizer
|
13 |
+
import math
|
14 |
+
import torch
|
15 |
+
import nltk
|
16 |
+
nltk.download('punkt')
|
17 |
|
18 |
docs = None
|
19 |
|
20 |
+
if torch.cuda.is_available():
|
21 |
+
dev = "cuda"
|
22 |
+
else:
|
23 |
+
dev = "cpu"
|
24 |
+
device = torch.device(dev)
|
25 |
|
26 |
def request_pathname(files):
|
27 |
if files is None:
|
28 |
return [[]]
|
29 |
return [[file.name, file.name.split('/')[-1]] for file in files]
|
30 |
|
31 |
+
def traducir_parrafos(parrafos, tokenizer, model, tam_bloque=8, ):
|
32 |
+
parrafos_traducidos = []
|
33 |
+
for parrafo in parrafos:
|
34 |
+
frases = sent_tokenize(parrafo)
|
35 |
+
batches = math.ceil(len(frases) / tam_bloque)
|
36 |
+
traducido = []
|
37 |
+
for i in range(batches):
|
38 |
+
|
39 |
+
bloque_enviado = frases[i*tam_bloque:(i+1)*tam_bloque]
|
40 |
+
model_inputs = tokenizer(bloque_enviado, return_tensors="pt",
|
41 |
+
padding=True, truncation=True,
|
42 |
+
max_length=500).to(device)
|
43 |
+
with torch.no_grad():
|
44 |
+
bloque_traducido = model.generate(**model_inputs)
|
45 |
+
traducido += bloque_traducido
|
46 |
+
traducido = [tokenizer.decode(t, skip_special_tokens=True) for t in traducido]
|
47 |
+
parrafos_traducidos += [" ".join(traducido)]
|
48 |
+
return parrafos_traducidos
|
49 |
+
|
50 |
+
def traducir_es_en(texto):
|
51 |
+
mname = "Helsinki-NLP/opus-mt-es-en"
|
52 |
+
tokenizer = MarianTokenizer.from_pretrained(mname)
|
53 |
+
model = MarianMTModel.from_pretrained(mname)
|
54 |
+
model.to(device)
|
55 |
+
|
56 |
+
lt = LineTokenizer()
|
57 |
+
batch_size = 8
|
58 |
+
parrafos = lt.tokenize(text_long)
|
59 |
+
par_tra = traducir_parrafos(parrafos, tokenizer, model)
|
60 |
+
return "\n".join(par_tra)
|
61 |
+
|
62 |
+
def traducir_en_es(texto):
|
63 |
+
mname = "Helsinki-NLP/opus-mt-en-es"
|
64 |
+
tokenizer = MarianTokenizer.from_pretrained(mname)
|
65 |
+
model = MarianMTModel.from_pretrained(mname)
|
66 |
+
model.to(device)
|
67 |
+
|
68 |
+
lt = LineTokenizer()
|
69 |
+
batch_size = 8
|
70 |
+
parrafos = lt.tokenize(text_long)
|
71 |
+
par_tra = traducir_parrafos(parrafos, tokenizer, model)
|
72 |
+
return "\n".join(par_tra)
|
73 |
|
74 |
def validate_dataset(dataset):
|
75 |
global docs
|
|
|
81 |
return "⚠️Esperando documentos..."
|
82 |
|
83 |
def do_ask(question, button, dataset):
|
|
|
84 |
global docs
|
85 |
docs_ready = dataset.iloc[-1, 0] != ""
|
86 |
if button == "✨Listo✨" and docs_ready:
|
87 |
for _, row in dataset.iterrows():
|
88 |
path = row['filepath']
|
89 |
text = Path(f'{path}').read_text()
|
90 |
+
text_en = traducir_es_en(text)
|
91 |
question_answerer = pipeline("question-answering", model='distilbert-base-cased-distilled-squad')
|
92 |
QA_input = {
|
93 |
+
'question': traducir_es_en(question),
|
94 |
+
'context': text_en
|
95 |
}
|
96 |
+
return traducir_en_es(question_answerer(QA_input)['answer'])
|
97 |
else:
|
98 |
return ""
|
99 |
|