Simon Salmon commited on
Commit
41d0dc8
·
1 Parent(s): e1cd7be

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +28 -16
app.py CHANGED
@@ -1,25 +1,37 @@
1
  import torch
2
- from transformers import T5ForConditionalGeneration,T5Tokenizer, AutoTokenizer, AutoModelForSeq2SeqLM, PegasusTokenizer, PegasusForConditionalGeneration
3
  import streamlit as st
 
 
 
4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5
  st.title("Auto Translate (To English)")
6
  text = st.text_input("Okay")
7
  st.text("What you wrote: ")
8
  st.write(text)
9
  st.text("English Translation: ")
10
-
11
- model_name = 'tuner007/pegasus_paraphrase'
12
- tokenizer = PegasusTokenizer.from_pretrained(model_name)
13
- model = PegasusForConditionalGeneration.from_pretrained(model_name)
14
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
15
- model = model.to(device)
16
-
17
- def get_response(text):
18
- batch = tokenizer([text],truncation=True,padding='longest',max_length=60, return_tensors="pt").to(device)
19
- translated = model.generate(**batch,max_length=60, do_sample=True, num_return_sequences=10, temperature=1.5)
20
- tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
21
- return tgt_text
22
-
23
  if text:
24
- translated_text = get_response(text)
25
- st.write(translated_text if translated_text else "No translation found")
 
1
  import torch
2
+ from transformers import T5ForConditionalGeneration,T5Tokenizer, AutoTokenizer, AutoModelForSeq2SeqLM
3
  import streamlit as st
4
+ model_name = st.text_input("Pick a Model", "seduerr/t5-pawraphrase")
5
+ tokenizer = AutoTokenizer.from_pretrained(model_name)
6
+ model = AutoModelForSeq2SeqLM.from_pretrained("t5-base")
7
 
8
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
9
+ model = model.to(device)
10
+ def translate_to_english(model, tokenizer, text):
11
+ translated_text = []
12
+ text = "paraphrase: " + text + " </s>"
13
+ encoding = tokenizer.encode_plus(text,pad_to_max_length=True, return_tensors="pt")
14
+ input_ids, attention_masks = encoding["input_ids"].to(device), encoding["attention_mask"].to(device)
15
+ beam_outputs = model.generate(
16
+ input_ids=input_ids, attention_mask=attention_masks,
17
+ do_sample=True,
18
+ max_length=256,
19
+ top_k=120,
20
+ top_p=0.98,
21
+ early_stopping=True,
22
+ num_return_sequences=10
23
+ )
24
+ for beam_output in beam_outputs:
25
+ sent = tokenizer.decode(beam_output, skip_special_tokens=True,clean_up_tokenization_spaces=True)
26
+ print(sent)
27
+ translated_text.append(sent)
28
+ return translated_text
29
+
30
  st.title("Auto Translate (To English)")
31
  text = st.text_input("Okay")
32
  st.text("What you wrote: ")
33
  st.write(text)
34
  st.text("English Translation: ")
 
 
 
 
 
 
 
 
 
 
 
 
 
35
  if text:
36
+ translated_text = translate_to_english(model, tokenizer, text)
37
+ st.write(translated_text if translated_text else "No translation found")