Spaces:
Build error
Build error
Commit
•
d1f4023
1
Parent(s):
f00dd8a
Setting up
Browse files
README.md
CHANGED
@@ -1,12 +1,16 @@
|
|
1 |
-
---
|
2 |
-
title:
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
title: Text_to_Text_Translator
|
3 |
+
app_file: app.py
|
4 |
+
sdk: gradio
|
5 |
+
sdk_version: 4.32.0
|
6 |
+
---
|
7 |
+
## Text to Text Translator
|
8 |
+
|
9 |
+
Built a Text to Text Translator using NLTK and Transformers.
|
10 |
+
- Supports Translation of English to Bengali, Tamil, Telugu, Gujarati, Marathi and Hindi.
|
11 |
+
- Uses BanglaT5 which achieved an exceptional score of <b>25.2</b> on SacreBLEU metric while mt5 (Industry Standard) scored much lower at <b>22.5</b>
|
12 |
+
|
13 |
+
|
14 |
+
Future Work:
|
15 |
+
- Adding functionality of uploading Images and Files
|
16 |
+
- OCR will run on these files and provide translation automatically
|
app.py
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#User Interface
|
2 |
+
|
3 |
+
import gradio as gr
|
4 |
+
import main
|
5 |
+
|
6 |
+
def test(text, src, dest):
|
7 |
+
ans = main.main_translation(text,dest,src)
|
8 |
+
return ans['output']
|
9 |
+
demo = gr.Interface(
|
10 |
+
test,
|
11 |
+
["textbox",
|
12 |
+
gr.Dropdown(
|
13 |
+
[("English", "en_XX"), ("Hindi","hi_IN"), ("Bengali","bn_IN"), ("Gujarati","gu_IN"), ("Tamil","ta_IN"), ("Telugu","te_IN"), ("Marathi","mr_IN")], label="Source", info="Select the Source Language!"
|
14 |
+
),
|
15 |
+
gr.Dropdown(
|
16 |
+
[("English", "en_XX"), ("Hindi","hi_IN"), ("Bengali","bn_IN"), ("Gujarati","gu_IN"), ("Tamil","ta_IN"), ("Telugu","te_IN"), ("Marathi","mr_IN")], label="Destination", info="Select the Destination Language!"
|
17 |
+
),
|
18 |
+
],
|
19 |
+
outputs=["textbox"],
|
20 |
+
)
|
21 |
+
|
22 |
+
demo.launch(share=True)
|
main.py
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import requests
|
2 |
+
from transformers import pipeline
|
3 |
+
import nltk
|
4 |
+
from nltk import sent_tokenize
|
5 |
+
from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
|
6 |
+
from transformers import pipeline
|
7 |
+
|
8 |
+
# nltk.download('punkt') # Run only once
|
9 |
+
|
10 |
+
tokenizer = MBart50TokenizerFast.from_pretrained("SnypzZz/Llama2-13b-Language-translate", src_lang="en_XX")
|
11 |
+
#pipe = pipeline("text2text-generation", model="SnypzZz/Llama2-13b-Language-translate", tokenizer=tokenizer)
|
12 |
+
model = None
|
13 |
+
model_loaded = False
|
14 |
+
|
15 |
+
api_token_header = ""
|
16 |
+
with open('./secret.py', 'r') as f:
|
17 |
+
api_token_header = f.read()
|
18 |
+
|
19 |
+
def load_model():
|
20 |
+
global model, model_loaded
|
21 |
+
model = MBartForConditionalGeneration.from_pretrained("SnypzZz/Llama2-13b-Language-translate")
|
22 |
+
model_loaded =True
|
23 |
+
return model
|
24 |
+
|
25 |
+
def translation(text,dest_lang,dest_lang_code, src_lang_code):
|
26 |
+
|
27 |
+
if(dest_lang_code == src_lang_code):
|
28 |
+
return "Please select different languages to translate between."
|
29 |
+
|
30 |
+
# headers = {"Authorization": f"Bearer {secrets_sih.api_token_header}"}
|
31 |
+
headers = {"Authorization": f"Bearer {api_token_header}"}
|
32 |
+
|
33 |
+
# Bengali Done
|
34 |
+
if(dest_lang == "Bengali" and src_lang_code == "en_XX"):
|
35 |
+
API_URL = "https://api-inference.huggingface.co/models/csebuetnlp/banglat5_nmt_en_bn"
|
36 |
+
def query(payload):
|
37 |
+
response = requests.post(API_URL, headers=headers, json=payload)
|
38 |
+
return response.json()
|
39 |
+
output = query({
|
40 |
+
"inputs": text,
|
41 |
+
})
|
42 |
+
print(output)
|
43 |
+
return output[0]['translation_text']
|
44 |
+
else:
|
45 |
+
global model
|
46 |
+
if model:
|
47 |
+
pass
|
48 |
+
else:
|
49 |
+
model = load_model()
|
50 |
+
loaded_model = model
|
51 |
+
tokenizer = MBart50TokenizerFast.from_pretrained("SnypzZz/Llama2-13b-Language-translate", src_lang=src_lang_code)
|
52 |
+
#model_inputs = tokenizer(text, return_tensors="pt")
|
53 |
+
loaded_model_inputs = tokenizer(text, return_tensors="pt")
|
54 |
+
|
55 |
+
# translate
|
56 |
+
generated_tokens = loaded_model.generate(
|
57 |
+
**loaded_model_inputs,
|
58 |
+
forced_bos_token_id=tokenizer.lang_code_to_id[dest_lang_code]
|
59 |
+
)
|
60 |
+
output = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
|
61 |
+
print(output)
|
62 |
+
return output[0]
|
63 |
+
|
64 |
+
def main_translation(text,dest_lang_code,src_lang_code):
|
65 |
+
|
66 |
+
codes = {"en_XX":"English","bn_IN":"Bengali", "en_GB":"English","gu_IN":"Gujarati","hi_IN":"Hindi","ta_IN":"Tamil","te_IN":"Telugu","mr_IN":"Marathi"}
|
67 |
+
dest_lang = codes[dest_lang_code]
|
68 |
+
src_lang = codes[src_lang_code]
|
69 |
+
|
70 |
+
sentences = sent_tokenize(text)
|
71 |
+
output = ""
|
72 |
+
for line in sentences:
|
73 |
+
output += translation(line,dest_lang,dest_lang_code, src_lang_code)
|
74 |
+
return {"output":output}
|
75 |
+
|
76 |
+
|
77 |
+
print(main_translation("hello world", "hi_IN", "en_XX"))
|