- app.py +79 -4
- requirements.txt +3 -0
app.py
CHANGED
|
@@ -1,7 +1,82 @@
|
|
| 1 |
import gradio as gr
|
|
|
|
| 2 |
|
| 3 |
-
|
| 4 |
-
|
| 5 |
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
from gliner import GLiNER
|
| 3 |
|
| 4 |
+
# Load the model outside of the processing function to ensure it's only loaded once
|
| 5 |
+
model = GLiNER.from_pretrained("nvidia/gliner-pii")
|
| 6 |
|
| 7 |
+
# Define a comprehensive default list of PII labels for the user
|
| 8 |
+
DEFAULT_LABELS = [
|
| 9 |
+
"person_name", "email", "phone_number", "address",
|
| 10 |
+
"ssn", "credit_card_number", "date", "user_name"
|
| 11 |
+
]
|
| 12 |
+
|
| 13 |
+
def redact_pii(text, labels_str, threshold):
|
| 14 |
+
"""
|
| 15 |
+
Redact PII from text using the GLiNER model
|
| 16 |
+
|
| 17 |
+
Args:
|
| 18 |
+
text (str): Input text to redact
|
| 19 |
+
labels_str (str): Comma-separated string of labels or empty for defaults
|
| 20 |
+
threshold (float): Confidence threshold for entity detection
|
| 21 |
+
|
| 22 |
+
Returns:
|
| 23 |
+
str: Redacted text with PII replaced by <REDACTED>
|
| 24 |
+
"""
|
| 25 |
+
# Process labels
|
| 26 |
+
if not labels_str or not labels_str.strip():
|
| 27 |
+
labels = DEFAULT_LABELS
|
| 28 |
+
else:
|
| 29 |
+
labels = [label.strip() for label in labels_str.split(",")]
|
| 30 |
+
|
| 31 |
+
# Predict entities
|
| 32 |
+
entities = model.predict_entities(text, labels, threshold=threshold)
|
| 33 |
+
|
| 34 |
+
# Redact entities in reverse order to prevent index shifting
|
| 35 |
+
redacted_text = text
|
| 36 |
+
for entity in reversed(entities):
|
| 37 |
+
start = entity["start"]
|
| 38 |
+
end = entity["end"]
|
| 39 |
+
redacted_text = redacted_text[:start] + "<REDACTED>" + redacted_text[end:]
|
| 40 |
+
|
| 41 |
+
return redacted_text
|
| 42 |
+
|
| 43 |
+
# Create the Gradio interface
|
| 44 |
+
with gr.Blocks(title="GLiNER PII Redaction App") as demo:
|
| 45 |
+
gr.Markdown("# GLiNER PII Redaction App")
|
| 46 |
+
gr.Markdown("This app uses the NVIDIA GLiNER PII model to detect and redact personally identifiable information from text.")
|
| 47 |
+
|
| 48 |
+
with gr.Row():
|
| 49 |
+
with gr.Column():
|
| 50 |
+
input_text = gr.Textbox(
|
| 51 |
+
label="Input Text (Plain, Markdown, or HTML)",
|
| 52 |
+
lines=10,
|
| 53 |
+
placeholder="Enter the text you want to redact..."
|
| 54 |
+
)
|
| 55 |
+
custom_labels = gr.Textbox(
|
| 56 |
+
label="Custom PII Labels (comma-separated)",
|
| 57 |
+
placeholder="person_name, email, phone_number, address, ssn, credit_card_number, date, user_name"
|
| 58 |
+
)
|
| 59 |
+
threshold = gr.Slider(
|
| 60 |
+
minimum=0.0,
|
| 61 |
+
maximum=1.0,
|
| 62 |
+
step=0.05,
|
| 63 |
+
value=0.5,
|
| 64 |
+
label="Confidence Threshold"
|
| 65 |
+
)
|
| 66 |
+
submit_btn = gr.Button("Redact PII")
|
| 67 |
+
|
| 68 |
+
with gr.Column():
|
| 69 |
+
output_text = gr.Textbox(
|
| 70 |
+
label="Redacted Output",
|
| 71 |
+
lines=10,
|
| 72 |
+
interactive=False
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
submit_btn.click(
|
| 76 |
+
fn=redact_pii,
|
| 77 |
+
inputs=[input_text, custom_labels, threshold],
|
| 78 |
+
outputs=output_text
|
| 79 |
+
)
|
| 80 |
+
|
| 81 |
+
if __name__ == "__main__":
|
| 82 |
+
demo.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
gradio
|
| 2 |
+
gliner
|
| 3 |
+
transformers
|