Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,80 +1,73 @@
|
|
| 1 |
import os
|
| 2 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "0"
|
| 3 |
|
| 4 |
-
from transformers import
|
| 5 |
-
MBart50TokenizerFast,
|
| 6 |
-
MBartForConditionalGeneration,
|
| 7 |
-
AutoConfig
|
| 8 |
-
)
|
| 9 |
import gradio as gr
|
| 10 |
|
| 11 |
-
# ---- Load model
|
| 12 |
model_name = "Mudasir692/mbart-eng-ur"
|
| 13 |
|
|
|
|
| 14 |
config = AutoConfig.from_pretrained(model_name)
|
| 15 |
-
|
| 16 |
-
# Fix missing or invalid parameters
|
| 17 |
-
if getattr(config, "early_stopping", None) is None:
|
| 18 |
config.early_stopping = True
|
| 19 |
-
if getattr(config, "max_length", None) is None:
|
| 20 |
-
config.max_length = 128 # β
set a safe limit
|
| 21 |
-
if getattr(config, "num_beams", None) is None:
|
| 22 |
-
config.num_beams = 4
|
| 23 |
|
| 24 |
-
tokenizer = MBart50TokenizerFast.from_pretrained(model_name
|
| 25 |
model = MBartForConditionalGeneration.from_pretrained(model_name, config=config)
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
# ---- Translation function ----
|
| 28 |
-
def
|
| 29 |
if not text.strip():
|
| 30 |
return "Please enter some English text."
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
inputs = tokenizer(text, return_tensors="pt", padding=True)
|
| 33 |
-
translated_tokens = model.generate(
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
num_beams=4,
|
| 37 |
-
early_stopping=True
|
| 38 |
-
)
|
| 39 |
-
urdu_output = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
|
| 40 |
-
return urdu_output
|
| 41 |
|
| 42 |
-
# ----
|
| 43 |
examples = [
|
| 44 |
-
["How are you?"],
|
| 45 |
-
["
|
| 46 |
-
["
|
| 47 |
-
["I am learning Artificial Intelligence."],
|
| 48 |
-
["Thank you very much!"]
|
| 49 |
]
|
| 50 |
|
|
|
|
| 51 |
app = gr.Interface(
|
| 52 |
-
fn=
|
| 53 |
-
inputs=
|
| 54 |
-
|
|
|
|
|
|
|
|
|
|
| 55 |
examples=examples,
|
| 56 |
-
title="
|
| 57 |
description="""
|
| 58 |
<div style='text-align:center;'>
|
| 59 |
-
<h3>Translate English sentences into Urdu using a fine-tuned mBART model.</h3>
|
| 60 |
<p style='color:gray;'>Built by <b>Khurram Basharat</b> β powered by Transformers & Gradio.</p>
|
| 61 |
-
<p><i>β³
|
| 62 |
</div>
|
| 63 |
""",
|
| 64 |
-
theme="soft",
|
| 65 |
css="""
|
| 66 |
body {
|
| 67 |
-
background: linear-gradient(
|
| 68 |
font-family: 'Segoe UI', sans-serif;
|
| 69 |
-
background-attachment: fixed;
|
| 70 |
}
|
| 71 |
.gr-button-primary {
|
| 72 |
background-color: #1e3799 !important;
|
| 73 |
color: white !important;
|
| 74 |
-
border-radius: 8px !important;
|
| 75 |
-
}
|
| 76 |
-
h1, h3 {
|
| 77 |
-
color: #0a3d62 !important;
|
| 78 |
}
|
| 79 |
""",
|
| 80 |
)
|
|
|
|
| 1 |
import os
|
| 2 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "0"
|
| 3 |
|
| 4 |
+
from transformers import MBart50TokenizerFast, MBartForConditionalGeneration, AutoConfig
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
import gradio as gr
|
| 6 |
|
| 7 |
+
# ---- Load model & tokenizer ----
|
| 8 |
model_name = "Mudasir692/mbart-eng-ur"
|
| 9 |
|
| 10 |
+
# Fix config issue
|
| 11 |
config = AutoConfig.from_pretrained(model_name)
|
| 12 |
+
if config.early_stopping is None:
|
|
|
|
|
|
|
| 13 |
config.early_stopping = True
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
+
tokenizer = MBart50TokenizerFast.from_pretrained(model_name)
|
| 16 |
model = MBartForConditionalGeneration.from_pretrained(model_name, config=config)
|
| 17 |
|
| 18 |
+
# ---- Language mapping ----
|
| 19 |
+
LANG_CODES = {
|
| 20 |
+
"Urdu": "ur_PK",
|
| 21 |
+
"Arabic": "ar_AR",
|
| 22 |
+
"Hindi": "hi_IN",
|
| 23 |
+
}
|
| 24 |
+
|
| 25 |
# ---- Translation function ----
|
| 26 |
+
def translate_text(text, target_lang):
|
| 27 |
if not text.strip():
|
| 28 |
return "Please enter some English text."
|
| 29 |
|
| 30 |
+
tgt_lang_code = LANG_CODES.get(target_lang, "ur_PK")
|
| 31 |
+
tokenizer.src_lang = "en_XX"
|
| 32 |
+
tokenizer.tgt_lang = tgt_lang_code
|
| 33 |
+
|
| 34 |
inputs = tokenizer(text, return_tensors="pt", padding=True)
|
| 35 |
+
translated_tokens = model.generate(**inputs)
|
| 36 |
+
output = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
|
| 37 |
+
return output
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
+
# ---- Examples ----
|
| 40 |
examples = [
|
| 41 |
+
["How are you?", "Urdu"],
|
| 42 |
+
["Where are you going?", "Arabic"],
|
| 43 |
+
["This is my new project.", "Hindi"],
|
|
|
|
|
|
|
| 44 |
]
|
| 45 |
|
| 46 |
+
# ---- Gradio Interface ----
|
| 47 |
app = gr.Interface(
|
| 48 |
+
fn=translate_text,
|
| 49 |
+
inputs=[
|
| 50 |
+
gr.Textbox(label="Enter English Text", placeholder="Type your English sentence here...", lines=2),
|
| 51 |
+
gr.Dropdown(["Urdu", "Arabic", "Hindi"], label="Select Target Language", value="Urdu")
|
| 52 |
+
],
|
| 53 |
+
outputs=gr.Textbox(label="Translation", lines=2),
|
| 54 |
examples=examples,
|
| 55 |
+
title="π Multi-Language Translator",
|
| 56 |
description="""
|
| 57 |
<div style='text-align:center;'>
|
| 58 |
+
<h3>Translate English sentences into Urdu, Arabic, or Hindi using a fine-tuned mBART model.</h3>
|
| 59 |
<p style='color:gray;'>Built by <b>Khurram Basharat</b> β powered by Transformers & Gradio.</p>
|
| 60 |
+
<p><i>β³ The model loads on first use, please wait a few seconds.</i></p>
|
| 61 |
</div>
|
| 62 |
""",
|
|
|
|
| 63 |
css="""
|
| 64 |
body {
|
| 65 |
+
background: linear-gradient(to bottom right, #f1f2f6, #dff9fb);
|
| 66 |
font-family: 'Segoe UI', sans-serif;
|
|
|
|
| 67 |
}
|
| 68 |
.gr-button-primary {
|
| 69 |
background-color: #1e3799 !important;
|
| 70 |
color: white !important;
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
}
|
| 72 |
""",
|
| 73 |
)
|