Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,25 +2,18 @@ import gradio as gr
|
|
| 2 |
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
|
| 3 |
|
| 4 |
# --- CONFIGURATION ---
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
MODEL_H2K_REPO = "ankitklakra/hindi-to-kurukh"
|
| 8 |
|
| 9 |
# --- LOAD RESOURCES ---
|
| 10 |
-
# 1. Load the "Dictionary" (Tokenizer) from Google
|
| 11 |
-
|
| 12 |
print("Loading Tokenizer...")
|
| 13 |
tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
|
| 14 |
|
| 15 |
-
|
| 16 |
-
print("Loading Kurukh -> Hindi Model...")
|
| 17 |
model_k2h = AutoModelForSeq2SeqLM.from_pretrained(MODEL_K2H_REPO)
|
| 18 |
-
|
| 19 |
-
print("Loading Hindi -> Kurukh Model...")
|
| 20 |
model_h2k = AutoModelForSeq2SeqLM.from_pretrained(MODEL_H2K_REPO)
|
| 21 |
|
| 22 |
-
#
|
| 23 |
-
|
| 24 |
pipe_k2h = pipeline("text2text-generation", model=model_k2h, tokenizer=tokenizer)
|
| 25 |
pipe_h2k = pipeline("text2text-generation", model=model_h2k, tokenizer=tokenizer)
|
| 26 |
|
|
@@ -29,26 +22,20 @@ def translate_text(text, direction):
|
|
| 29 |
if not text:
|
| 30 |
return ""
|
| 31 |
|
| 32 |
-
|
| 33 |
-
if direction == "Kurukh -> Hindi":
|
| 34 |
-
target_pipeline = pipe_k2h
|
| 35 |
-
else:
|
| 36 |
-
target_pipeline = pipe_h2k
|
| 37 |
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
|
|
|
|
|
|
| 41 |
|
| 42 |
-
# --- THE
|
| 43 |
-
|
|
|
|
| 44 |
|
| 45 |
-
gr.Markdown(
|
| 46 |
-
|
| 47 |
-
# 🇮🇳 AI Kurukh (Kurux) Translator
|
| 48 |
-
### Preserving Tribal Languages with Artificial Intelligence
|
| 49 |
-
*Powered by Custom Fine-Tuned Google mT5 Models*
|
| 50 |
-
"""
|
| 51 |
-
)
|
| 52 |
|
| 53 |
with gr.Row():
|
| 54 |
direction = gr.Radio(
|
|
@@ -58,21 +45,11 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 58 |
)
|
| 59 |
|
| 60 |
with gr.Row():
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
label="Input Text",
|
| 64 |
-
placeholder="Type your sentence here...",
|
| 65 |
-
lines=5
|
| 66 |
-
)
|
| 67 |
-
translate_btn = gr.Button("Translate 🚀", variant="primary")
|
| 68 |
-
|
| 69 |
-
with gr.Column():
|
| 70 |
-
output_text = gr.Textbox(
|
| 71 |
-
label="Translation Result",
|
| 72 |
-
lines=5,
|
| 73 |
-
show_copy_button=True
|
| 74 |
-
)
|
| 75 |
|
|
|
|
|
|
|
| 76 |
gr.Examples(
|
| 77 |
examples=[
|
| 78 |
["निघै नामे इन्द्रा हिकै?", "Kurukh -> Hindi"],
|
|
@@ -80,11 +57,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
| 80 |
["तुम्हारा नाम क्या है?", "Hindi -> Kurukh"],
|
| 81 |
["मुझे पानी दो।", "Hindi -> Kurukh"]
|
| 82 |
],
|
| 83 |
-
inputs=[input_text, direction]
|
| 84 |
-
label="Click on an example to test:"
|
| 85 |
)
|
| 86 |
|
| 87 |
translate_btn.click(fn=translate_text, inputs=[input_text, direction], outputs=output_text)
|
| 88 |
|
| 89 |
-
# Launch
|
| 90 |
demo.launch()
|
|
|
|
| 2 |
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
|
| 3 |
|
| 4 |
# --- CONFIGURATION ---
|
| 5 |
+
MODEL_K2H_REPO = "ankitklakra/kurukh-to-hindi"
|
| 6 |
+
MODEL_H2K_REPO = "ankitklakra/hindi-to-kurukh"
|
|
|
|
| 7 |
|
| 8 |
# --- LOAD RESOURCES ---
|
|
|
|
|
|
|
| 9 |
print("Loading Tokenizer...")
|
| 10 |
tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
|
| 11 |
|
| 12 |
+
print("Loading Models...")
|
|
|
|
| 13 |
model_k2h = AutoModelForSeq2SeqLM.from_pretrained(MODEL_K2H_REPO)
|
|
|
|
|
|
|
| 14 |
model_h2k = AutoModelForSeq2SeqLM.from_pretrained(MODEL_H2K_REPO)
|
| 15 |
|
| 16 |
+
# Create Pipelines
|
|
|
|
| 17 |
pipe_k2h = pipeline("text2text-generation", model=model_k2h, tokenizer=tokenizer)
|
| 18 |
pipe_h2k = pipeline("text2text-generation", model=model_h2k, tokenizer=tokenizer)
|
| 19 |
|
|
|
|
| 22 |
if not text:
|
| 23 |
return ""
|
| 24 |
|
| 25 |
+
target_pipeline = pipe_k2h if direction == "Kurukh -> Hindi" else pipe_h2k
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
+
try:
|
| 28 |
+
results = target_pipeline(text, max_length=128)
|
| 29 |
+
return results[0]['generated_text']
|
| 30 |
+
except Exception as e:
|
| 31 |
+
return f"Error: {str(e)}"
|
| 32 |
|
| 33 |
+
# --- THE UI ---
|
| 34 |
+
|
| 35 |
+
with gr.Blocks() as demo:
|
| 36 |
|
| 37 |
+
gr.Markdown("# 🇮🇳 AI Kurukh (Kurux) Translator")
|
| 38 |
+
gr.Markdown("### Preserving Tribal Languages with Artificial Intelligence")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
with gr.Row():
|
| 41 |
direction = gr.Radio(
|
|
|
|
| 45 |
)
|
| 46 |
|
| 47 |
with gr.Row():
|
| 48 |
+
input_text = gr.Textbox(label="Input Text", lines=5, placeholder="Type here...")
|
| 49 |
+
output_text = gr.Textbox(label="Translation Result", lines=5)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
+
translate_btn = gr.Button("Translate 🚀", variant="primary")
|
| 52 |
+
|
| 53 |
gr.Examples(
|
| 54 |
examples=[
|
| 55 |
["निघै नामे इन्द्रा हिकै?", "Kurukh -> Hindi"],
|
|
|
|
| 57 |
["तुम्हारा नाम क्या है?", "Hindi -> Kurukh"],
|
| 58 |
["मुझे पानी दो।", "Hindi -> Kurukh"]
|
| 59 |
],
|
| 60 |
+
inputs=[input_text, direction]
|
|
|
|
| 61 |
)
|
| 62 |
|
| 63 |
translate_btn.click(fn=translate_text, inputs=[input_text, direction], outputs=output_text)
|
| 64 |
|
|
|
|
| 65 |
demo.launch()
|