Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -2,21 +2,24 @@ import re
|
|
2 |
import gradio as gr
|
3 |
from gliner import GLiNER
|
4 |
from cerberus import Validator
|
|
|
5 |
|
6 |
# ----------------------------------------------------------------------------
|
7 |
# Load model + labels
|
8 |
# ----------------------------------------------------------------------------
|
9 |
|
10 |
model = GLiNER.from_pretrained("urchade/gliner_multi_pii-v1")
|
|
|
11 |
|
12 |
with open("labels.txt", "r", encoding="utf-8") as f:
|
13 |
labels = [line.strip() for line in f.readlines()]
|
14 |
|
|
|
|
|
15 |
# ----------------------------------------------------------------------------
|
16 |
# Simple Cerberus validation for incoming data
|
17 |
# ----------------------------------------------------------------------------
|
18 |
|
19 |
-
# We expect a dict with at least {"text": "<some string>"}
|
20 |
schema = {
|
21 |
"text": {
|
22 |
"type": "string",
|
@@ -26,74 +29,88 @@ schema = {
|
|
26 |
|
27 |
validator = Validator(schema)
|
28 |
|
29 |
-
|
30 |
def validate_input(data: dict) -> str:
|
31 |
-
"""Validate that data has a non-empty 'text' key."""
|
32 |
if not validator.validate(data):
|
33 |
-
# If invalid, raise an exception. You could handle this more gracefully if you like.
|
34 |
raise ValueError(f"Invalid input data. Errors: {validator.errors}")
|
35 |
return data["text"]
|
36 |
|
37 |
# ----------------------------------------------------------------------------
|
38 |
-
#
|
39 |
# ----------------------------------------------------------------------------
|
40 |
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
for entity in entities:
|
57 |
-
label = entity['label'].replace(" ", "_").upper()
|
58 |
-
original_text = entity['text']
|
59 |
-
start_idx, end_idx = entity['start'], entity['end']
|
60 |
-
|
61 |
-
if label not in entity_map:
|
62 |
-
entity_map[label] = [original_text]
|
63 |
-
idx = 1
|
64 |
else:
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
else:
|
69 |
-
|
70 |
-
|
|
|
|
|
|
|
71 |
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
anonymized_text += f"<PII_{label}_{idx}>"
|
76 |
-
next_start = end_idx
|
77 |
|
78 |
-
|
79 |
-
|
80 |
-
return anonymized_text, entity_map
|
81 |
|
|
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
with their original strings from entity_map.
|
87 |
-
"""
|
88 |
|
|
|
89 |
def replace_match(match):
|
90 |
-
label = match.group(1)
|
91 |
-
idx_str = match.group(2)
|
92 |
-
idx = int(idx_str) - 1
|
93 |
-
|
94 |
if label in entity_map and 0 <= idx < len(entity_map[label]):
|
95 |
return entity_map[label][idx]
|
96 |
-
return match.group(0)
|
97 |
|
98 |
pattern = r"<PII_(\w+)_(\d+)>"
|
99 |
return re.sub(pattern, replace_match, anonymized_response)
|
@@ -103,18 +120,15 @@ def deanonymize_text(anonymized_response, entity_map):
|
|
103 |
# ----------------------------------------------------------------------------
|
104 |
|
105 |
def anonymize_fn(original_text):
|
106 |
-
# We’ll do a simple dict so we can pass it to our Cerberus validator:
|
107 |
data = {"text": original_text}
|
108 |
try:
|
109 |
user_text = validate_input(data)
|
110 |
except ValueError as e:
|
111 |
-
# If invalid, show error in Gradio output
|
112 |
return "", {}, f"Validation error: {str(e)}"
|
113 |
|
114 |
-
anonymized, entities =
|
115 |
return anonymized, entities, "Успешно анонимизировано!"
|
116 |
|
117 |
-
|
118 |
def deanonymize_fn(anonymized_llm_response, entity_map):
|
119 |
if not anonymized_llm_response.strip():
|
120 |
return "", "Вставьте анонимизированный текст."
|
@@ -124,11 +138,9 @@ def deanonymize_fn(anonymized_llm_response, entity_map):
|
|
124 |
result = deanonymize_text(anonymized_llm_response, entity_map)
|
125 |
return result, "Успешно деанонимизировано!"
|
126 |
|
127 |
-
|
128 |
md_text = """# Анонимизатор психотерапевтических сессий
|
129 |
|
130 |
-
Вставьте текст в раздел "Исходный
|
131 |
-
|
132 |
"""
|
133 |
|
134 |
with gr.Blocks() as demo:
|
@@ -144,9 +156,7 @@ with gr.Blocks() as demo:
|
|
144 |
)
|
145 |
button_anon = gr.Button("Анонимизировать")
|
146 |
|
147 |
-
# Hidden state to store the entity map
|
148 |
entity_map_state = gr.State()
|
149 |
-
|
150 |
message_out = gr.Textbox(label="Status", interactive=False)
|
151 |
|
152 |
button_anon.click(
|
@@ -173,4 +183,4 @@ with gr.Blocks() as demo:
|
|
173 |
)
|
174 |
|
175 |
if __name__ == "__main__":
|
176 |
-
demo.launch()
|
|
|
2 |
import gradio as gr
|
3 |
from gliner import GLiNER
|
4 |
from cerberus import Validator
|
5 |
+
from transformers import AutoTokenizer
|
6 |
|
7 |
# ----------------------------------------------------------------------------
|
8 |
# Load model + labels
|
9 |
# ----------------------------------------------------------------------------
|
10 |
|
11 |
model = GLiNER.from_pretrained("urchade/gliner_multi_pii-v1")
|
12 |
+
tokenizer = AutoTokenizer.from_pretrained("xlm-roberta-base")
|
13 |
|
14 |
with open("labels.txt", "r", encoding="utf-8") as f:
|
15 |
labels = [line.strip() for line in f.readlines()]
|
16 |
|
17 |
+
MAX_TOKENS = 512 # безопасный лимит токенов на один фрагмент
|
18 |
+
|
19 |
# ----------------------------------------------------------------------------
|
20 |
# Simple Cerberus validation for incoming data
|
21 |
# ----------------------------------------------------------------------------
|
22 |
|
|
|
23 |
schema = {
|
24 |
"text": {
|
25 |
"type": "string",
|
|
|
29 |
|
30 |
validator = Validator(schema)
|
31 |
|
|
|
32 |
def validate_input(data: dict) -> str:
|
|
|
33 |
if not validator.validate(data):
|
|
|
34 |
raise ValueError(f"Invalid input data. Errors: {validator.errors}")
|
35 |
return data["text"]
|
36 |
|
37 |
# ----------------------------------------------------------------------------
|
38 |
+
# Chunking + Anonymization logic
|
39 |
# ----------------------------------------------------------------------------
|
40 |
|
41 |
+
def split_text_into_chunks(text, max_tokens=MAX_TOKENS):
|
42 |
+
words = text.split()
|
43 |
+
chunks = []
|
44 |
+
chunk = []
|
45 |
+
chunk_token_count = 0
|
46 |
+
current_offset = 0
|
47 |
+
|
48 |
+
for word in words:
|
49 |
+
token_count = len(tokenizer.tokenize(word))
|
50 |
+
if chunk_token_count + token_count > max_tokens:
|
51 |
+
chunk_text = ' '.join(chunk)
|
52 |
+
chunks.append((chunk_text, current_offset))
|
53 |
+
current_offset += len(chunk_text) + 1
|
54 |
+
chunk = [word]
|
55 |
+
chunk_token_count = token_count
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
else:
|
57 |
+
chunk.append(word)
|
58 |
+
chunk_token_count += token_count
|
59 |
+
|
60 |
+
if chunk:
|
61 |
+
chunk_text = ' '.join(chunk)
|
62 |
+
chunks.append((chunk_text, current_offset))
|
63 |
+
|
64 |
+
return chunks
|
65 |
+
|
66 |
+
def anonymize_text_long(text):
|
67 |
+
chunks = split_text_into_chunks(text)
|
68 |
+
full_anonymized = ""
|
69 |
+
global_entity_map = {}
|
70 |
+
|
71 |
+
for chunk_text, _ in chunks:
|
72 |
+
entities = model.predict_entities(chunk_text, labels=labels, threshold=0.2)
|
73 |
+
entities.sort(key=lambda e: e['start'])
|
74 |
+
|
75 |
+
anonymized_chunk = ""
|
76 |
+
next_start = 0
|
77 |
+
|
78 |
+
for entity in entities:
|
79 |
+
label = entity['label'].replace(" ", "_").upper()
|
80 |
+
original_text = entity['text']
|
81 |
+
start_idx, end_idx = entity['start'], entity['end']
|
82 |
+
|
83 |
+
if label not in global_entity_map:
|
84 |
+
global_entity_map[label] = [original_text]
|
85 |
+
idx = 1
|
86 |
else:
|
87 |
+
if original_text in global_entity_map[label]:
|
88 |
+
idx = global_entity_map[label].index(original_text) + 1
|
89 |
+
else:
|
90 |
+
global_entity_map[label].append(original_text)
|
91 |
+
idx = len(global_entity_map[label])
|
92 |
|
93 |
+
anonymized_chunk += chunk_text[next_start:start_idx]
|
94 |
+
anonymized_chunk += f"<PII_{label}_{idx}>"
|
95 |
+
next_start = end_idx
|
|
|
|
|
96 |
|
97 |
+
anonymized_chunk += chunk_text[next_start:]
|
98 |
+
full_anonymized += anonymized_chunk + " "
|
|
|
99 |
|
100 |
+
return full_anonymized.strip(), global_entity_map
|
101 |
|
102 |
+
# ----------------------------------------------------------------------------
|
103 |
+
# De-anonymization logic
|
104 |
+
# ----------------------------------------------------------------------------
|
|
|
|
|
105 |
|
106 |
+
def deanonymize_text(anonymized_response, entity_map):
|
107 |
def replace_match(match):
|
108 |
+
label = match.group(1)
|
109 |
+
idx_str = match.group(2)
|
110 |
+
idx = int(idx_str) - 1
|
|
|
111 |
if label in entity_map and 0 <= idx < len(entity_map[label]):
|
112 |
return entity_map[label][idx]
|
113 |
+
return match.group(0)
|
114 |
|
115 |
pattern = r"<PII_(\w+)_(\d+)>"
|
116 |
return re.sub(pattern, replace_match, anonymized_response)
|
|
|
120 |
# ----------------------------------------------------------------------------
|
121 |
|
122 |
def anonymize_fn(original_text):
|
|
|
123 |
data = {"text": original_text}
|
124 |
try:
|
125 |
user_text = validate_input(data)
|
126 |
except ValueError as e:
|
|
|
127 |
return "", {}, f"Validation error: {str(e)}"
|
128 |
|
129 |
+
anonymized, entities = anonymize_text_long(user_text)
|
130 |
return anonymized, entities, "Успешно анонимизировано!"
|
131 |
|
|
|
132 |
def deanonymize_fn(anonymized_llm_response, entity_map):
|
133 |
if not anonymized_llm_response.strip():
|
134 |
return "", "Вставьте анонимизированный текст."
|
|
|
138 |
result = deanonymize_text(anonymized_llm_response, entity_map)
|
139 |
return result, "Успешно деанонимизировано!"
|
140 |
|
|
|
141 |
md_text = """# Анонимизатор психотерапевтических сессий
|
142 |
|
143 |
+
Вставьте текст в раздел \"Исходный текст\", чтобы анонимизировать сензитивные данные.
|
|
|
144 |
"""
|
145 |
|
146 |
with gr.Blocks() as demo:
|
|
|
156 |
)
|
157 |
button_anon = gr.Button("Анонимизировать")
|
158 |
|
|
|
159 |
entity_map_state = gr.State()
|
|
|
160 |
message_out = gr.Textbox(label="Status", interactive=False)
|
161 |
|
162 |
button_anon.click(
|
|
|
183 |
)
|
184 |
|
185 |
if __name__ == "__main__":
|
186 |
+
demo.launch()
|