Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -6,26 +6,37 @@ from transformers import MarianMTModel, MarianTokenizer
|
|
6 |
import os
|
7 |
os.system("pip install sentencepiece")
|
8 |
|
9 |
-
#
|
10 |
-
|
|
|
|
|
|
|
11 |
tokenizer_en_to_ur = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ur")
|
12 |
|
13 |
-
model_ur_to_en = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-ur-en")
|
14 |
tokenizer_ur_to_en = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ur-en")
|
15 |
|
|
|
|
|
|
|
|
|
|
|
16 |
# Function to translate text
|
17 |
def translate(text, direction):
|
|
|
|
|
|
|
18 |
if direction == "English to Urdu":
|
19 |
tokenizer, model = tokenizer_en_to_ur, model_en_to_ur
|
20 |
else:
|
21 |
tokenizer, model = tokenizer_ur_to_en, model_ur_to_en
|
22 |
|
23 |
-
# Tokenize input text
|
24 |
-
inputs = tokenizer(text, return_tensors="pt", padding=
|
25 |
|
26 |
# Generate translation
|
27 |
with torch.no_grad():
|
28 |
-
translated = model.generate(**inputs)
|
29 |
|
30 |
# Decode output text
|
31 |
output = tokenizer.decode(translated[0], skip_special_tokens=True)
|
@@ -39,8 +50,8 @@ interface = gr.Interface(
|
|
39 |
gr.Radio(["English to Urdu", "Urdu to English"], label="Translation Direction", value="English to Urdu"),
|
40 |
],
|
41 |
outputs=gr.Textbox(label="Translated Text"),
|
42 |
-
title="English ↔ Urdu Translator",
|
43 |
-
description="Translate text between English and Urdu using a neural machine translation model.",
|
44 |
)
|
45 |
|
46 |
# Launch the Gradio app
|
|
|
6 |
import os
|
7 |
os.system("pip install sentencepiece")
|
8 |
|
9 |
+
# Check if GPU is available and use it
|
10 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
11 |
+
|
12 |
+
# Load models and tokenizers once (globally)
|
13 |
+
model_en_to_ur = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-en-ur").to(device)
|
14 |
tokenizer_en_to_ur = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ur")
|
15 |
|
16 |
+
model_ur_to_en = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-ur-en").to(device)
|
17 |
tokenizer_ur_to_en = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ur-en")
|
18 |
|
19 |
+
# Apply torch.compile() for optimization (if using PyTorch 2.0+)
|
20 |
+
if torch.__version__ >= "2.0":
|
21 |
+
model_en_to_ur = torch.compile(model_en_to_ur)
|
22 |
+
model_ur_to_en = torch.compile(model_ur_to_en)
|
23 |
+
|
24 |
# Function to translate text
|
25 |
def translate(text, direction):
|
26 |
+
if not text.strip():
|
27 |
+
return "Please enter some text to translate."
|
28 |
+
|
29 |
if direction == "English to Urdu":
|
30 |
tokenizer, model = tokenizer_en_to_ur, model_en_to_ur
|
31 |
else:
|
32 |
tokenizer, model = tokenizer_ur_to_en, model_ur_to_en
|
33 |
|
34 |
+
# Tokenize input text (optimized padding)
|
35 |
+
inputs = tokenizer(text, return_tensors="pt", padding="longest", truncation=True).to(device)
|
36 |
|
37 |
# Generate translation
|
38 |
with torch.no_grad():
|
39 |
+
translated = model.generate(**inputs, max_length=512)
|
40 |
|
41 |
# Decode output text
|
42 |
output = tokenizer.decode(translated[0], skip_special_tokens=True)
|
|
|
50 |
gr.Radio(["English to Urdu", "Urdu to English"], label="Translation Direction", value="English to Urdu"),
|
51 |
],
|
52 |
outputs=gr.Textbox(label="Translated Text"),
|
53 |
+
title="⚡ Fast English ↔ Urdu Translator",
|
54 |
+
description="Translate text between English and Urdu quickly using a neural machine translation model with GPU acceleration.",
|
55 |
)
|
56 |
|
57 |
# Launch the Gradio app
|