Spaces:
Running
Running
refactoring 2
Browse files- generation.py +120 -0
- initialize.py +84 -0
- interface.py +171 -0
- main.py +30 -501
- search.py +36 -0
- utils.py +55 -0
generation.py
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from enum import Enum
|
3 |
+
from openai import OpenAI
|
4 |
+
import google.generativeai as genai
|
5 |
+
from llama_index.core.llms import ChatMessage
|
6 |
+
from prompts import LEGAL_POSITION_PROMPT, SYSTEM_PROMPT
|
7 |
+
|
8 |
+
|
9 |
+
class GenerationProvider(str, Enum):
|
10 |
+
OPENAI = "openai"
|
11 |
+
GEMINI = "gemini"
|
12 |
+
|
13 |
+
|
14 |
+
class GenerationModelName(str, Enum):
|
15 |
+
# OpenAI models
|
16 |
+
GPT4_LEGAL = "ft:gpt-4o-mini-2024-07-18:personal:legal-position-1500:Aaiu4WZd"
|
17 |
+
# Gemini models
|
18 |
+
GEMINI_FLASH = "gemini-1.5-flash"
|
19 |
+
|
20 |
+
|
21 |
+
# Schema for OpenAI response
|
22 |
+
LEGAL_POSITION_SCHEMA = {
|
23 |
+
"type": "json_schema",
|
24 |
+
"json_schema": {
|
25 |
+
"name": "lp_schema",
|
26 |
+
"schema": {
|
27 |
+
"type": "object",
|
28 |
+
"properties": {
|
29 |
+
"title": {"type": "string", "description": "Title of the legal position"},
|
30 |
+
"text": {"type": "string", "description": "Text of the legal position"},
|
31 |
+
"proceeding": {"type": "string", "description": "Type of court proceedings"},
|
32 |
+
"category": {"type": "string", "description": "Category of the legal position"},
|
33 |
+
},
|
34 |
+
"required": ["title", "text", "proceeding", "category"],
|
35 |
+
"additionalProperties": False
|
36 |
+
},
|
37 |
+
"strict": True
|
38 |
+
}
|
39 |
+
}
|
40 |
+
|
41 |
+
|
42 |
+
def generate_legal_position(court_decision_text: str, comment_input: str, provider: str, model_name: str) -> dict:
|
43 |
+
if not isinstance(court_decision_text, str) or not court_decision_text.strip():
|
44 |
+
return {
|
45 |
+
"title": "Invalid input",
|
46 |
+
"text": "Court decision text is required and must be non-empty.",
|
47 |
+
"proceeding": "Error",
|
48 |
+
"category": "Error"
|
49 |
+
}
|
50 |
+
|
51 |
+
try:
|
52 |
+
content = LEGAL_POSITION_PROMPT.format(
|
53 |
+
court_decision_text=court_decision_text,
|
54 |
+
comment=comment_input if comment_input else "Коментар відсутній"
|
55 |
+
)
|
56 |
+
|
57 |
+
if provider == GenerationProvider.OPENAI.value:
|
58 |
+
client = OpenAI()
|
59 |
+
response = client.chat.completions.create(
|
60 |
+
model=model_name,
|
61 |
+
messages=[
|
62 |
+
{"role": "system", "content": SYSTEM_PROMPT},
|
63 |
+
{"role": "user", "content": content}
|
64 |
+
],
|
65 |
+
response_format={"type": "json_object"},
|
66 |
+
temperature=0
|
67 |
+
)
|
68 |
+
parsed_response = json.loads(response.choices[0].message.content)
|
69 |
+
|
70 |
+
# Перевірка та конвертація полів
|
71 |
+
if 'text_lp' in parsed_response and 'text' not in parsed_response:
|
72 |
+
parsed_response['text'] = parsed_response.pop('text_lp')
|
73 |
+
|
74 |
+
elif provider == GenerationProvider.GEMINI.value:
|
75 |
+
generation_config = {
|
76 |
+
"temperature": 0,
|
77 |
+
"max_output_tokens": 8192,
|
78 |
+
"response_mime_type": "application/json",
|
79 |
+
}
|
80 |
+
|
81 |
+
model = genai.GenerativeModel(
|
82 |
+
model_name=model_name,
|
83 |
+
generation_config=generation_config,
|
84 |
+
)
|
85 |
+
|
86 |
+
chat = model.start_chat(history=[])
|
87 |
+
response = chat.send_message(
|
88 |
+
f"{SYSTEM_PROMPT}\n\n{content}",
|
89 |
+
)
|
90 |
+
parsed_response = json.loads(response.text)
|
91 |
+
|
92 |
+
# Та сама перевірка для Gemini
|
93 |
+
if 'text_lp' in parsed_response and 'text' not in parsed_response:
|
94 |
+
parsed_response['text'] = parsed_response.pop('text_lp')
|
95 |
+
|
96 |
+
else:
|
97 |
+
raise ValueError(f"Unsupported provider: {provider}")
|
98 |
+
|
99 |
+
# Валідація результату
|
100 |
+
required_fields = ["title", "text", "proceeding", "category"]
|
101 |
+
if all(field in parsed_response for field in required_fields):
|
102 |
+
return parsed_response
|
103 |
+
|
104 |
+
missing_fields = [field for field in required_fields if field not in parsed_response]
|
105 |
+
raise ValueError(f"Missing required fields: {', '.join(missing_fields)}")
|
106 |
+
|
107 |
+
except json.JSONDecodeError as e:
|
108 |
+
return {
|
109 |
+
"title": "Error parsing response",
|
110 |
+
"text": f"Failed to parse JSON response: {str(e)}",
|
111 |
+
"proceeding": "Error",
|
112 |
+
"category": "Error"
|
113 |
+
}
|
114 |
+
except Exception as e:
|
115 |
+
return {
|
116 |
+
"title": str(parsed_response.get('title', 'Error')),
|
117 |
+
"text": str(parsed_response.get('text_lp', parsed_response.get('text', str(e)))),
|
118 |
+
"proceeding": str(parsed_response.get('proceeding', 'Error')),
|
119 |
+
"category": str(parsed_response.get('category', 'Error'))
|
120 |
+
}
|
initialize.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import sys
|
2 |
+
import boto3
|
3 |
+
from pathlib import Path
|
4 |
+
from llama_index.core import Settings
|
5 |
+
from llama_index.core.storage.docstore import SimpleDocumentStore
|
6 |
+
from llama_index.retrievers.bm25 import BM25Retriever
|
7 |
+
from llama_index.core.retrievers import QueryFusionRetriever
|
8 |
+
|
9 |
+
from config import aws_access_key_id, aws_secret_access_key
|
10 |
+
|
11 |
+
|
12 |
+
class AppState:
|
13 |
+
_instance = None
|
14 |
+
retriever_bm25 = None
|
15 |
+
|
16 |
+
def __new__(cls):
|
17 |
+
if cls._instance is None:
|
18 |
+
cls._instance = super(AppState, cls).__new__(cls)
|
19 |
+
return cls._instance
|
20 |
+
|
21 |
+
|
22 |
+
# Параметри S3
|
23 |
+
BUCKET_NAME = "legal-position"
|
24 |
+
PREFIX_RETRIEVER = "Save_Index/"
|
25 |
+
LOCAL_DIR = Path("Save_Index_Local")
|
26 |
+
|
27 |
+
# Створюємо глобальний екземпляр стану
|
28 |
+
app_state = AppState()
|
29 |
+
|
30 |
+
|
31 |
+
def initialize_s3_client():
|
32 |
+
return boto3.client(
|
33 |
+
"s3",
|
34 |
+
aws_access_key_id=aws_access_key_id,
|
35 |
+
aws_secret_access_key=aws_secret_access_key,
|
36 |
+
region_name="eu-north-1"
|
37 |
+
)
|
38 |
+
|
39 |
+
|
40 |
+
def download_s3_file(s3_client, bucket_name, s3_key, local_path):
|
41 |
+
s3_client.download_file(bucket_name, s3_key, str(local_path))
|
42 |
+
print(f"Завантажено: {s3_key} -> {local_path}")
|
43 |
+
|
44 |
+
|
45 |
+
def download_s3_folder(s3_client, bucket_name, prefix, local_dir):
|
46 |
+
response = s3_client.list_objects_v2(Bucket=bucket_name, Prefix=prefix)
|
47 |
+
if 'Contents' in response:
|
48 |
+
for obj in response['Contents']:
|
49 |
+
s3_key = obj['Key']
|
50 |
+
if s3_key.endswith('/'):
|
51 |
+
continue
|
52 |
+
local_file_path = local_dir / Path(s3_key).relative_to(prefix)
|
53 |
+
local_file_path.parent.mkdir(parents=True, exist_ok=True)
|
54 |
+
s3_client.download_file(bucket_name, s3_key, str(local_file_path))
|
55 |
+
print(f"Завантажено: {s3_key} -> {local_file_path}")
|
56 |
+
|
57 |
+
|
58 |
+
def initialize_components():
|
59 |
+
try:
|
60 |
+
persist_path = Path("Save_Index_Local")
|
61 |
+
|
62 |
+
if not persist_path.exists():
|
63 |
+
raise FileNotFoundError(f"Directory not found: {persist_path}")
|
64 |
+
|
65 |
+
required_files = ['docstore_es_filter.json', 'bm25_retriever_es']
|
66 |
+
missing_files = [f for f in required_files if not (persist_path / f).exists()]
|
67 |
+
|
68 |
+
if missing_files:
|
69 |
+
raise FileNotFoundError(f"Missing required files: {', '.join(missing_files)}")
|
70 |
+
|
71 |
+
docstore = SimpleDocumentStore.from_persist_path(str(persist_path / "docstore_es_filter.json"))
|
72 |
+
bm25_retriever = BM25Retriever.from_persist_dir(str(persist_path / "bm25_retriever_es"))
|
73 |
+
|
74 |
+
# Зберігаємо retriever_bm25 в глобальному стані
|
75 |
+
app_state.retriever_bm25 = QueryFusionRetriever(
|
76 |
+
[bm25_retriever],
|
77 |
+
similarity_top_k=Settings.similarity_top_k,
|
78 |
+
num_queries=1,
|
79 |
+
use_async=True,
|
80 |
+
)
|
81 |
+
return True
|
82 |
+
except Exception as e:
|
83 |
+
print(f"Error initializing components: {str(e)}", file=sys.stderr)
|
84 |
+
return False
|
interface.py
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from typing import List
|
3 |
+
import json
|
4 |
+
from enum import Enum
|
5 |
+
|
6 |
+
from analysis import ModelProvider, ModelName, PrecedentAnalysisWorkflow
|
7 |
+
from generation import GenerationProvider, GenerationModelName, generate_legal_position
|
8 |
+
from utils import extract_court_decision_text, get_links_html, get_links_html_lp
|
9 |
+
from search import search_with_ai_action
|
10 |
+
|
11 |
+
|
12 |
+
def create_gradio_interface():
|
13 |
+
def update_generation_model_choices(provider):
|
14 |
+
if provider == GenerationProvider.OPENAI.value:
|
15 |
+
return gr.Dropdown(choices=[m.value for m in GenerationModelName if m.value.startswith("ft")])
|
16 |
+
else:
|
17 |
+
return gr.Dropdown(choices=[m.value for m in GenerationModelName if m.value.startswith("gemini")])
|
18 |
+
|
19 |
+
def update_analysis_model_choices(provider):
|
20 |
+
if provider == ModelProvider.OPENAI.value:
|
21 |
+
return gr.Dropdown(choices=[m.value for m in ModelName if m.value.startswith("gpt")])
|
22 |
+
else:
|
23 |
+
return gr.Dropdown(choices=[m.value for m in ModelName if m.value.startswith("claude")])
|
24 |
+
|
25 |
+
async def generate_position_action(url, provider, model_name, comment_input):
|
26 |
+
try:
|
27 |
+
court_decision_text = extract_court_decision_text(url)
|
28 |
+
legal_position_json = generate_legal_position(court_decision_text, comment_input, provider, model_name)
|
29 |
+
position_output_content = (
|
30 |
+
f"**Короткий зміст позиції суду за введеним рішенням (модель: {model_name}):**\n"
|
31 |
+
f"*{legal_position_json['title']}*: \n"
|
32 |
+
f"{legal_position_json['text']} "
|
33 |
+
f"**Категорія:** \n{legal_position_json['category']} "
|
34 |
+
f"({legal_position_json['proceeding']})\n\n"
|
35 |
+
)
|
36 |
+
return position_output_content, legal_position_json
|
37 |
+
except Exception as e:
|
38 |
+
return f"Error during position generation: {str(e)}", None
|
39 |
+
|
40 |
+
async def analyze_action(legal_position_json, question, nodes, provider, model_name):
|
41 |
+
try:
|
42 |
+
workflow = PrecedentAnalysisWorkflow(
|
43 |
+
provider=ModelProvider(provider),
|
44 |
+
model_name=ModelName(model_name)
|
45 |
+
)
|
46 |
+
|
47 |
+
query = (
|
48 |
+
f"{legal_position_json['title']}: "
|
49 |
+
f"{legal_position_json['text']}: "
|
50 |
+
f"{legal_position_json['proceeding']}: "
|
51 |
+
f"{legal_position_json['category']}"
|
52 |
+
)
|
53 |
+
|
54 |
+
response_text = await workflow.run(
|
55 |
+
query=query,
|
56 |
+
question=question,
|
57 |
+
nodes=nodes
|
58 |
+
)
|
59 |
+
|
60 |
+
output = f"**Аналіз ШІ (модель: {model_name}):**\n{response_text}\n\n"
|
61 |
+
output += "**Наявні в базі Правові Позицій Верховного Суду:**\n\n"
|
62 |
+
|
63 |
+
analysis_lines = response_text.split('\n')
|
64 |
+
for line in analysis_lines:
|
65 |
+
if line.startswith('* ['):
|
66 |
+
index = line[3:line.index(']')]
|
67 |
+
node = nodes[int(index) - 1]
|
68 |
+
source_node = node.node
|
69 |
+
|
70 |
+
source_title = source_node.metadata.get('title', 'Невідомий заголовок')
|
71 |
+
source_text_lp = node.text
|
72 |
+
doc_ids = source_node.metadata.get('doc_id')
|
73 |
+
lp_id = source_node.metadata.get('lp_id')
|
74 |
+
|
75 |
+
links = get_links_html(doc_ids)
|
76 |
+
links_lp = get_links_html_lp(lp_id)
|
77 |
+
|
78 |
+
output += f"[{index}]: *{source_title}* | {source_text_lp} | {links_lp} | {links}\n\n"
|
79 |
+
|
80 |
+
return output
|
81 |
+
|
82 |
+
except Exception as e:
|
83 |
+
return f"Error during analysis: {str(e)}"
|
84 |
+
|
85 |
+
with gr.Blocks() as app:
|
86 |
+
gr.Markdown("# Аналізатор релевантних Правових Позицій Верховного Суду для нового судового рішення")
|
87 |
+
|
88 |
+
with gr.Row():
|
89 |
+
comment_input = gr.Textbox(label="Коментар до формування короткого змісту судового рішення:")
|
90 |
+
url_input = gr.Textbox(label="URL судового рішення:")
|
91 |
+
question_input = gr.Textbox(label="Уточнююче питання для аналізу:")
|
92 |
+
|
93 |
+
with gr.Row():
|
94 |
+
# Провайдер для генерування
|
95 |
+
generation_provider_dropdown = gr.Dropdown(
|
96 |
+
choices=[p.value for p in GenerationProvider],
|
97 |
+
value=GenerationProvider.GEMINI.value,
|
98 |
+
label="Провайдер AI для генерування",
|
99 |
+
)
|
100 |
+
generation_model_dropdown = gr.Dropdown(
|
101 |
+
choices=[m.value for m in GenerationModelName if m.value.startswith("gemini")],
|
102 |
+
value=GenerationModelName.GEMINI_FLASH.value,
|
103 |
+
label="Модель для генерування",
|
104 |
+
)
|
105 |
+
|
106 |
+
with gr.Row():
|
107 |
+
# Пр��вайдер для аналізу
|
108 |
+
analysis_provider_dropdown = gr.Dropdown(
|
109 |
+
choices=[p.value for p in ModelProvider],
|
110 |
+
value=ModelProvider.OPENAI.value,
|
111 |
+
label="Провайдер AI для аналізу",
|
112 |
+
)
|
113 |
+
analysis_model_dropdown = gr.Dropdown(
|
114 |
+
choices=[m.value for m in ModelName if m.value.startswith("gpt")],
|
115 |
+
value=ModelName.GPT4o_MINI.value,
|
116 |
+
label="Модель для аналізу",
|
117 |
+
)
|
118 |
+
|
119 |
+
with gr.Row():
|
120 |
+
generate_position_button = gr.Button("Генерувати короткий зміст позиції суду")
|
121 |
+
search_with_ai_button = gr.Button("Пошук", interactive=False)
|
122 |
+
analyze_button = gr.Button("Аналіз", interactive=False)
|
123 |
+
|
124 |
+
position_output = gr.Markdown(label="Короткий зміст позиції суду за введеним рішенням")
|
125 |
+
search_output = gr.Markdown(label="Результат пошуку")
|
126 |
+
analysis_output = gr.Markdown(label="Результат аналізу")
|
127 |
+
|
128 |
+
state_lp_json = gr.State()
|
129 |
+
state_nodes = gr.State()
|
130 |
+
|
131 |
+
# Підключення функцій до кнопок та подій
|
132 |
+
generate_position_button.click(
|
133 |
+
fn=generate_position_action,
|
134 |
+
inputs=[url_input, generation_provider_dropdown, generation_model_dropdown, comment_input],
|
135 |
+
outputs=[position_output, state_lp_json]
|
136 |
+
).then(
|
137 |
+
fn=lambda: gr.update(interactive=True),
|
138 |
+
inputs=None,
|
139 |
+
outputs=search_with_ai_button
|
140 |
+
)
|
141 |
+
|
142 |
+
search_with_ai_button.click(
|
143 |
+
fn=search_with_ai_action,
|
144 |
+
inputs=state_lp_json,
|
145 |
+
outputs=[search_output, state_nodes]
|
146 |
+
).then(
|
147 |
+
fn=lambda: gr.update(interactive=True),
|
148 |
+
inputs=None,
|
149 |
+
outputs=analyze_button
|
150 |
+
)
|
151 |
+
|
152 |
+
analyze_button.click(
|
153 |
+
fn=analyze_action,
|
154 |
+
inputs=[state_lp_json, question_input, state_nodes, analysis_provider_dropdown, analysis_model_dropdown],
|
155 |
+
outputs=analysis_output
|
156 |
+
)
|
157 |
+
|
158 |
+
# Оновлення списків моделей при зміні провайдера
|
159 |
+
generation_provider_dropdown.change(
|
160 |
+
fn=update_generation_model_choices,
|
161 |
+
inputs=generation_provider_dropdown,
|
162 |
+
outputs=generation_model_dropdown
|
163 |
+
)
|
164 |
+
|
165 |
+
analysis_provider_dropdown.change(
|
166 |
+
fn=update_analysis_model_choices,
|
167 |
+
inputs=analysis_provider_dropdown,
|
168 |
+
outputs=analysis_model_dropdown
|
169 |
+
)
|
170 |
+
|
171 |
+
return app
|
main.py
CHANGED
@@ -1,512 +1,41 @@
|
|
1 |
-
import os
|
2 |
-
import re
|
3 |
-
import gradio as gr
|
4 |
-
import pandas as pd
|
5 |
-
import requests
|
6 |
-
import json
|
7 |
-
import faiss
|
8 |
-
import nest_asyncio
|
9 |
import sys
|
10 |
-
import
|
11 |
-
|
12 |
from pathlib import Path
|
13 |
-
from
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
StorageContext,
|
21 |
-
ServiceContext,
|
22 |
-
VectorStoreIndex,
|
23 |
-
Settings,
|
24 |
-
load_index_from_storage
|
25 |
-
)
|
26 |
-
from llama_index.llms.openai import OpenAI
|
27 |
-
from llama_index.core.llms import ChatMessage
|
28 |
-
from llama_index.core.schema import IndexNode
|
29 |
-
from llama_index.core.storage.docstore import SimpleDocumentStore
|
30 |
-
from llama_index.retrievers.bm25 import BM25Retriever
|
31 |
-
from llama_index.embeddings.openai import OpenAIEmbedding
|
32 |
-
# from llama_index.vector_stores.faiss import FaissVectorStore
|
33 |
-
from llama_index.core.retrievers import QueryFusionRetriever
|
34 |
-
from llama_index.core.workflow import Event, Context, Workflow, StartEvent, StopEvent, step
|
35 |
-
from llama_index.core.schema import NodeWithScore
|
36 |
-
from llama_index.core.prompts import PromptTemplate
|
37 |
-
from llama_index.core.response_synthesizers import ResponseMode, get_response_synthesizer
|
38 |
-
|
39 |
-
from config import embed_model, Settings, openai_api_key, anthropic_api_key, aws_access_key_id, aws_secret_access_key
|
40 |
-
from analysis import ModelProvider, ModelName, PrecedentAnalysisWorkflow
|
41 |
-
|
42 |
-
from prompts import SYSTEM_PROMPT, LEGAL_POSITION_PROMPT, PRECEDENT_ANALYSIS_TEMPLATE
|
43 |
-
|
44 |
-
|
45 |
-
# from dotenv import load_dotenv
|
46 |
-
#
|
47 |
-
# load_dotenv()
|
48 |
-
#
|
49 |
-
# aws_access_key_id = os.getenv("AWS_ACCESS_KEY_ID")
|
50 |
-
# aws_secret_access_key = os.getenv("AWS_SECRET_ACCESS_KEY")
|
51 |
-
# openai_api_key = os.getenv("OPENAI_API_KEY")
|
52 |
-
# anthropic_api_key=os.getenv("ANTHROPIC_API_KEY")
|
53 |
-
# genai.configure(api_key=os.environ["GOOGLE_API_KEY"])
|
54 |
-
#
|
55 |
-
#
|
56 |
-
# embed_model = OpenAIEmbedding(model_name="text-embedding-3-small")
|
57 |
-
# Settings.embed_model = embed_model
|
58 |
-
# Settings.context_window = 20000
|
59 |
-
# Settings.chunk_size = 2048
|
60 |
-
# Settings.similarity_top_k = 20
|
61 |
-
|
62 |
-
|
63 |
-
# Параметри S3
|
64 |
-
BUCKET_NAME = "legal-position"
|
65 |
-
PREFIX_RETRIEVER = "Save_Index/" # Префікс для всього вмісту, який потрібно завантажити
|
66 |
-
LOCAL_DIR = Path("Save_Index_Local") # Локальна директорія для збереження даних з S3
|
67 |
-
|
68 |
-
|
69 |
-
# Ініціалізація клієнта S3
|
70 |
-
s3_client = boto3.client(
|
71 |
-
"s3",
|
72 |
-
aws_access_key_id=aws_access_key_id,
|
73 |
-
aws_secret_access_key=aws_secret_access_key,
|
74 |
-
region_name="eu-north-1"
|
75 |
)
|
76 |
-
|
77 |
-
|
78 |
-
# Створюємо локальну директорію, якщо вона не існує
|
79 |
-
LOCAL_DIR.mkdir(parents=True, exist_ok=True)
|
80 |
-
|
81 |
-
# Функція для завантаження файлу з S3
|
82 |
-
def download_s3_file(bucket_name, s3_key, local_path):
|
83 |
-
s3_client.download_file(bucket_name, s3_key, str(local_path))
|
84 |
-
print(f"Завантажено: {s3_key} -> {local_path}")
|
85 |
-
|
86 |
-
# Функція для завантаження всієї папки з S3 у локальну директорію
|
87 |
-
def download_s3_folder(bucket_name, prefix, local_dir):
|
88 |
-
response = s3_client.list_objects_v2(Bucket=bucket_name, Prefix=prefix)
|
89 |
-
if 'Contents' in response:
|
90 |
-
for obj in response['Contents']:
|
91 |
-
s3_key = obj['Key']
|
92 |
-
# Пропускаємо "папку" (кореневий префікс) у S3
|
93 |
-
if s3_key.endswith('/'):
|
94 |
-
continue
|
95 |
-
# Визначаємо локальний шлях, де буде збережений файл
|
96 |
-
local_file_path = local_dir / Path(s3_key).relative_to(prefix)
|
97 |
-
local_file_path.parent.mkdir(parents=True, exist_ok=True) # створення підкаталогів, якщо потрібно
|
98 |
-
# Завантажуємо файл
|
99 |
-
s3_client.download_file(bucket_name, s3_key, str(local_file_path))
|
100 |
-
print(f"Завантажено: {s3_key} -> {local_file_path}")
|
101 |
-
|
102 |
-
# Перевіряємо, чи існує локальна директорія
|
103 |
-
if not LOCAL_DIR.exists():
|
104 |
-
print(f"Локальна директорія {LOCAL_DIR} відсутня. Починаємо завантаження...")
|
105 |
-
LOCAL_DIR.mkdir(parents=True, exist_ok=True) # Створення директорії
|
106 |
-
download_s3_folder(BUCKET_NAME, PREFIX_RETRIEVER, LOCAL_DIR)
|
107 |
-
else:
|
108 |
-
print(f"Локальна директорія {LOCAL_DIR} вже існує. Завантаження пропущено.")
|
109 |
-
|
110 |
-
|
111 |
|
112 |
# Apply nest_asyncio to handle nested async calls
|
113 |
nest_asyncio.apply()
|
114 |
|
115 |
-
|
116 |
-
nodes: list[NodeWithScore]
|
117 |
-
|
118 |
-
|
119 |
-
state_lp_json = gr.State()
|
120 |
-
state_nodes = gr.State()
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
def parse_doc_ids(doc_ids):
|
125 |
-
if doc_ids is None:
|
126 |
-
return []
|
127 |
-
if isinstance(doc_ids, list):
|
128 |
-
return [str(id).strip('[]') for id in doc_ids]
|
129 |
-
if isinstance(doc_ids, str):
|
130 |
-
cleaned = doc_ids.strip('[]').replace(' ', '')
|
131 |
-
if cleaned:
|
132 |
-
return [id.strip() for id in cleaned.split(',')]
|
133 |
-
return []
|
134 |
-
|
135 |
-
def get_links_html(doc_ids):
|
136 |
-
parsed_ids = parse_doc_ids(doc_ids)
|
137 |
-
if not parsed_ids:
|
138 |
-
return ""
|
139 |
-
links = [f"[Рішення ВС: {doc_id}](https://reyestr.court.gov.ua/Review/{doc_id})"
|
140 |
-
for doc_id in parsed_ids]
|
141 |
-
return ", ".join(links)
|
142 |
-
|
143 |
-
def parse_lp_ids(lp_ids):
|
144 |
-
if lp_ids is None:
|
145 |
-
return []
|
146 |
-
if isinstance(lp_ids, (str, int)):
|
147 |
-
cleaned = str(lp_ids).strip('[]').replace(' ', '')
|
148 |
-
if cleaned:
|
149 |
-
return [cleaned]
|
150 |
-
return []
|
151 |
-
|
152 |
-
def get_links_html_lp(lp_ids):
|
153 |
-
parsed_ids = parse_lp_ids(lp_ids)
|
154 |
-
if not parsed_ids:
|
155 |
-
return ""
|
156 |
-
links = [f"[ПП ВС: {lp_id}](https://lpd.court.gov.ua/home/search/{lp_id})" for lp_id in parsed_ids]
|
157 |
-
return ", ".join(links)
|
158 |
-
|
159 |
-
|
160 |
-
def initialize_components():
|
161 |
-
try:
|
162 |
-
# Використовуємо папку `Save_Index_Local`, куди завантажено файли з S3
|
163 |
-
persist_path = Path("Save_Index_Local")
|
164 |
-
|
165 |
-
# Перевірка існування локальної директорії
|
166 |
-
if not persist_path.exists():
|
167 |
-
raise FileNotFoundError(f"Directory not found: {persist_path}")
|
168 |
-
|
169 |
-
# Перевірка наявності необхідних файлів і папок
|
170 |
-
required_files = ['docstore_es_filter.json', 'bm25_retriever_es']
|
171 |
-
missing_files = [f for f in required_files if not (persist_path / f).exists()]
|
172 |
-
|
173 |
-
if missing_files:
|
174 |
-
raise FileNotFoundError(f"Missing required files: {', '.join(missing_files)}")
|
175 |
-
|
176 |
-
# Ініціалізація компонентів
|
177 |
-
global retriever_bm25
|
178 |
-
|
179 |
-
# Ініціалізація `SimpleDocumentStore` з `docstore_es_filter.json`
|
180 |
-
docstore = SimpleDocumentStore.from_persist_path(str(persist_path / "docstore_es_filter.json"))
|
181 |
-
|
182 |
-
# Ініціалізація `BM25Retriever` з папки `bm25_retriever_es`
|
183 |
-
bm25_retriever = BM25Retriever.from_persist_dir(str(persist_path / "bm25_retriever_es"))
|
184 |
-
|
185 |
-
# Ініціалізація `QueryFusionRetriever` з налаштуваннями
|
186 |
-
retriever_bm25 = QueryFusionRetriever(
|
187 |
-
[
|
188 |
-
bm25_retriever,
|
189 |
-
],
|
190 |
-
similarity_top_k=Settings.similarity_top_k,
|
191 |
-
num_queries=1,
|
192 |
-
use_async=True,
|
193 |
-
)
|
194 |
-
return True
|
195 |
-
except Exception as e:
|
196 |
-
print(f"Error initializing components: {str(e)}", file=sys.stderr)
|
197 |
-
return False
|
198 |
-
|
199 |
-
|
200 |
-
def extract_court_decision_text(url):
|
201 |
-
response = requests.get(url)
|
202 |
-
soup = BeautifulSoup(response.content, 'html.parser')
|
203 |
-
|
204 |
-
unwanted_texts = [
|
205 |
-
"Доступ до Реєстру здійснюється в тестовому (обмеженому) режимі.",
|
206 |
-
"З метою упередження перешкоджанню стабільній роботі Реєстру"
|
207 |
-
]
|
208 |
-
|
209 |
-
decision_text = ""
|
210 |
-
for paragraph in soup.find_all('p'):
|
211 |
-
text = paragraph.get_text(separator="\n").strip()
|
212 |
-
if not any(unwanted_text in text for unwanted_text in unwanted_texts):
|
213 |
-
decision_text += text + "\n"
|
214 |
-
return decision_text.strip()
|
215 |
-
|
216 |
-
|
217 |
-
# Constants for JSON schema
|
218 |
-
LEGAL_POSITION_SCHEMA = {
|
219 |
-
"type": "json_schema",
|
220 |
-
"json_schema": {
|
221 |
-
"name": "lp_schema",
|
222 |
-
"schema": {
|
223 |
-
"type": "object",
|
224 |
-
"properties": {
|
225 |
-
"title": {"type": "string", "description": "Title of the legal position"},
|
226 |
-
"text": {"type": "string", "description": "Text of the legal position"},
|
227 |
-
"proceeding": {"type": "string", "description": "Type of court proceedings"},
|
228 |
-
"category": {"type": "string", "description": "Category of the legal position"},
|
229 |
-
},
|
230 |
-
"required": ["title", "text", "proceeding", "category"],
|
231 |
-
"additionalProperties": False
|
232 |
-
},
|
233 |
-
"strict": True
|
234 |
-
}
|
235 |
-
}
|
236 |
-
|
237 |
-
|
238 |
-
# def generate_legal_position(court_decision_text, comment_input):
|
239 |
-
# try:
|
240 |
-
# # Ініціалізація моделі
|
241 |
-
# llm_lp = OpenAI(
|
242 |
-
# # model="ft:gpt-4o-mini-2024-07-18:personal:legal-position-400:AT3wvKsU",
|
243 |
-
# model="ft:gpt-4o-mini-2024-07-18:personal:legal-position-1500:Aaiu4WZd",
|
244 |
-
# temperature=0
|
245 |
-
# )
|
246 |
-
#
|
247 |
-
# # Формування повідомлень для чату
|
248 |
-
# # Формуємо контент з урахуванням коментаря
|
249 |
-
# content = LEGAL_POSITION_PROMPT.format(
|
250 |
-
# court_decision_text=court_decision_text,
|
251 |
-
# comment=comment_input if comment_input else "Коментар відсутній"
|
252 |
-
# )
|
253 |
-
#
|
254 |
-
# # Формування повідомлень д��я чату
|
255 |
-
# messages = [
|
256 |
-
# ChatMessage(role="system", content=SYSTEM_PROMPT),
|
257 |
-
# ChatMessage(role="user", content=content),
|
258 |
-
# ]
|
259 |
-
#
|
260 |
-
# # Отримання відповіді від моделі
|
261 |
-
# response = llm_lp.chat(messages, response_format=LEGAL_POSITION_SCHEMA)
|
262 |
-
#
|
263 |
-
# # Обробка відповіді
|
264 |
-
# parsed_response = json.loads(response.message.content)
|
265 |
-
#
|
266 |
-
# # Перевірка наявності обов'язкових полів
|
267 |
-
# if all(field in parsed_response for field in ["title", "text", "proceeding", "category"]):
|
268 |
-
# return parsed_response
|
269 |
-
#
|
270 |
-
# return {
|
271 |
-
# "title": "Error: Missing required fields in response",
|
272 |
-
# "text": response.message.content,
|
273 |
-
# "proceeding": "Unknown",
|
274 |
-
# "category": "Error"
|
275 |
-
# }
|
276 |
-
#
|
277 |
-
# except json.JSONDecodeError:
|
278 |
-
# return {
|
279 |
-
# "title": "Error parsing response",
|
280 |
-
# "text": response.message.content,
|
281 |
-
# "proceeding": "Unknown",
|
282 |
-
# "category": "Error"
|
283 |
-
# }
|
284 |
-
# except Exception as e:
|
285 |
-
# return {
|
286 |
-
# "title": "Unexpected error",
|
287 |
-
# "text": str(e),
|
288 |
-
# "proceeding": "Unknown",
|
289 |
-
# "category": "Error"
|
290 |
-
# }
|
291 |
-
|
292 |
-
|
293 |
-
def generate_legal_position(court_decision_text, comment_input):
|
294 |
-
if not isinstance(court_decision_text, str) or not court_decision_text.strip():
|
295 |
-
return {
|
296 |
-
"title": "Invalid input",
|
297 |
-
"text": "Court decision text is required and must be non-empty.",
|
298 |
-
"status": "Error"
|
299 |
-
}
|
300 |
-
|
301 |
try:
|
302 |
-
#
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
model = genai.GenerativeModel(
|
311 |
-
model_name="gemini-1.5-flash",
|
312 |
-
generation_config=generation_config,
|
313 |
-
system_instruction=SYSTEM_PROMPT,
|
314 |
-
)
|
315 |
-
|
316 |
-
content = LEGAL_POSITION_PROMPT.format(
|
317 |
-
court_decision_text=court_decision_text,
|
318 |
-
comment=comment_input if comment_input else "Коментар відсутній"
|
319 |
-
)
|
320 |
-
|
321 |
-
# Створення сесії чату
|
322 |
-
chat_session = model.start_chat(history=[])
|
323 |
-
|
324 |
-
response = chat_session.send_message(content)
|
325 |
-
|
326 |
-
# Обробка відповіді
|
327 |
-
parsed_response = json.loads(response.text)
|
328 |
-
|
329 |
-
# Перевірка наявності обов'язкових полів
|
330 |
-
if all(field in parsed_response for field in ["title", "text", "proceeding", "category"]):
|
331 |
-
return parsed_response
|
332 |
-
|
333 |
-
return {
|
334 |
-
"title": "Error: Missing required fields in response",
|
335 |
-
"text": response.text,
|
336 |
-
"proceeding": "Unknown",
|
337 |
-
"category": "Error"
|
338 |
-
}
|
339 |
-
|
340 |
-
except json.JSONDecodeError:
|
341 |
-
return {
|
342 |
-
"title": "Error parsing response",
|
343 |
-
"text": response.text,
|
344 |
-
"proceeding": "Unknown",
|
345 |
-
"category": "Error"
|
346 |
-
}
|
347 |
-
except Exception as e:
|
348 |
-
return {
|
349 |
-
"title": "Unexpected error",
|
350 |
-
"text": str(e),
|
351 |
-
"proceeding": "Unknown",
|
352 |
-
"category": "Error"
|
353 |
-
}
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
def create_gradio_interface():
|
359 |
-
async def generate_position_action(url):
|
360 |
-
try:
|
361 |
-
court_decision_text = extract_court_decision_text(url)
|
362 |
-
legal_position_json = generate_legal_position(court_decision_text, comment_input)
|
363 |
-
position_output_content = f"**Короткий зміст позиції суду за введеним рішенням:**\n *{legal_position_json['title']}*: \n{legal_position_json['text']} **Категорія:** \n{legal_position_json['category']} ({legal_position_json['proceeding']})\n\n"
|
364 |
-
return position_output_content, legal_position_json
|
365 |
-
except Exception as e:
|
366 |
-
return f"Error during position generation: {str(e)}", None
|
367 |
-
|
368 |
-
async def search_with_ai_action(legal_position_json):
|
369 |
-
try:
|
370 |
-
query_text = legal_position_json["title"] + ': ' + legal_position_json["text"] + ': ' + legal_position_json["proceeding"] + ': ' + legal_position_json["category"]
|
371 |
-
nodes = await retriever_bm25.aretrieve(query_text)
|
372 |
-
|
373 |
-
sources_output = "\n **Результати пошуку (наявні правові позиції ВСУ):** \n\n"
|
374 |
-
for index, node in enumerate(nodes, start=1):
|
375 |
-
source_title = node.node.metadata.get('title')
|
376 |
-
doc_ids = node.node.metadata.get('doc_id')
|
377 |
-
lp_ids = node.node.metadata.get('lp_id')
|
378 |
-
links = get_links_html(doc_ids)
|
379 |
-
links_lp = get_links_html_lp(lp_ids)
|
380 |
-
sources_output += f"\n[{index}] *{source_title}* {links_lp} 👉 Score: {node.score} {links}\n"
|
381 |
-
|
382 |
-
return sources_output, nodes
|
383 |
-
except Exception as e:
|
384 |
-
return f"Error during search: {str(e)}", None
|
385 |
-
|
386 |
-
async def analyze_action(legal_position_json, question, nodes, provider, model_name):
|
387 |
-
try:
|
388 |
-
workflow = PrecedentAnalysisWorkflow(
|
389 |
-
provider=ModelProvider(provider),
|
390 |
-
model_name=ModelName(model_name)
|
391 |
-
)
|
392 |
-
|
393 |
-
query = (
|
394 |
-
f"{legal_position_json['title']}: "
|
395 |
-
f"{legal_position_json['text']}: "
|
396 |
-
f"{legal_position_json['proceeding']}: "
|
397 |
-
f"{legal_position_json['category']}"
|
398 |
-
)
|
399 |
-
|
400 |
-
response_text = await workflow.run(
|
401 |
-
query=query,
|
402 |
-
question=question,
|
403 |
-
nodes=nodes
|
404 |
-
)
|
405 |
-
|
406 |
-
output = f"**Аналіз ШІ (модель: {model_name}):**\n{response_text}\n\n"
|
407 |
-
output += "**Наявні в базі Правові Позицій Верховного Суду:**\n\n"
|
408 |
-
|
409 |
-
analysis_lines = response_text.split('\n')
|
410 |
-
for line in analysis_lines:
|
411 |
-
if line.startswith('* ['):
|
412 |
-
index = line[3:line.index(']')]
|
413 |
-
node = nodes[int(index) - 1]
|
414 |
-
source_node = node.node
|
415 |
-
|
416 |
-
source_title = source_node.metadata.get('title', 'Невідомий заголовок')
|
417 |
-
source_text_lp = node.text
|
418 |
-
doc_ids = source_node.metadata.get('doc_id')
|
419 |
-
lp_id = source_node.metadata.get('lp_id')
|
420 |
-
|
421 |
-
links = get_links_html(doc_ids)
|
422 |
-
links_lp = get_links_html_lp(lp_id)
|
423 |
-
|
424 |
-
output += f"[{index}]: *{source_title}* | {source_text_lp} | {links_lp} | {links}\n\n"
|
425 |
-
|
426 |
-
return output
|
427 |
-
|
428 |
-
except Exception as e:
|
429 |
-
return f"Error during analysis: {str(e)}"
|
430 |
-
|
431 |
-
def update_model_choices(provider):
|
432 |
-
if provider == ModelProvider.OPENAI.value:
|
433 |
-
return gr.Dropdown(choices=[m.value for m in ModelName if m.value.startswith("gpt")])
|
434 |
else:
|
435 |
-
|
436 |
-
|
437 |
-
with gr.Blocks() as app:
|
438 |
-
# Далі ваш код інтерфейсу...
|
439 |
-
gr.Markdown("# Аналізатор релевантних Правових Позицій Верховного Суду для нового судового рішення")
|
440 |
-
|
441 |
-
with gr.Row():
|
442 |
-
comment_input = gr.Textbox(label="Коментар до формування короткого змісту судового рішення:")
|
443 |
-
url_input = gr.Textbox(label="URL судового рішення:")
|
444 |
-
question_input = gr.Textbox(label="Уточнююче питання для аналізу:")
|
445 |
|
446 |
-
|
447 |
-
|
448 |
-
|
449 |
-
|
450 |
-
|
451 |
-
|
452 |
-
|
453 |
-
|
454 |
-
value=ModelName.GPT4o_MINI.value,
|
455 |
-
label="Модель",
|
456 |
-
)
|
457 |
-
|
458 |
-
with gr.Row():
|
459 |
-
generate_position_button = gr.Button("Генерувати короткий зміст позиції суду")
|
460 |
-
search_with_ai_button = gr.Button("Пошук", interactive=False)
|
461 |
-
analyze_button = gr.Button("Аналіз", interactive=False)
|
462 |
-
|
463 |
-
position_output = gr.Markdown(label="Короткий зміст позиції суду за введеним рішенням")
|
464 |
-
search_output = gr.Markdown(label="Результат пошуку")
|
465 |
-
analysis_output = gr.Markdown(label="Результат аналізу")
|
466 |
-
|
467 |
-
state_lp_json = gr.State()
|
468 |
-
state_nodes = gr.State()
|
469 |
-
|
470 |
-
# Підключення функцій до кнопок
|
471 |
-
generate_position_button.click(
|
472 |
-
fn=generate_position_action,
|
473 |
-
inputs=url_input,
|
474 |
-
outputs=[position_output, state_lp_json]
|
475 |
-
).then(
|
476 |
-
fn=lambda: gr.update(interactive=True),
|
477 |
-
inputs=None,
|
478 |
-
outputs=search_with_ai_button
|
479 |
-
)
|
480 |
-
|
481 |
-
search_with_ai_button.click(
|
482 |
-
fn=search_with_ai_action,
|
483 |
-
inputs=state_lp_json,
|
484 |
-
outputs=[search_output, state_nodes]
|
485 |
-
).then(
|
486 |
-
fn=lambda: gr.update(interactive=True),
|
487 |
-
inputs=None,
|
488 |
-
outputs=analyze_button
|
489 |
-
)
|
490 |
-
|
491 |
-
analyze_button.click(
|
492 |
-
fn=analyze_action,
|
493 |
-
inputs=[state_lp_json, question_input, state_nodes, provider_dropdown, model_dropdown],
|
494 |
-
outputs=analysis_output
|
495 |
-
)
|
496 |
-
|
497 |
-
provider_dropdown.change(
|
498 |
-
fn=update_model_choices,
|
499 |
-
inputs=provider_dropdown,
|
500 |
-
outputs=model_dropdown
|
501 |
-
)
|
502 |
-
|
503 |
-
return app
|
504 |
|
505 |
-
|
506 |
-
|
507 |
-
|
508 |
-
app = create_gradio_interface()
|
509 |
-
app.launch(share=True)
|
510 |
-
else:
|
511 |
-
print("Failed to initialize components. Please check the paths and try again.", file=sys.stderr)
|
512 |
-
sys.exit(1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import sys
|
2 |
+
import nest_asyncio
|
|
|
3 |
from pathlib import Path
|
4 |
+
from initialize import (
|
5 |
+
initialize_components,
|
6 |
+
initialize_s3_client,
|
7 |
+
download_s3_folder,
|
8 |
+
LOCAL_DIR,
|
9 |
+
BUCKET_NAME,
|
10 |
+
PREFIX_RETRIEVER
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
)
|
12 |
+
from interface import create_gradio_interface
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
# Apply nest_asyncio to handle nested async calls
|
15 |
nest_asyncio.apply()
|
16 |
|
17 |
+
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
try:
|
19 |
+
# Створюємо локальну директорію
|
20 |
+
LOCAL_DIR.mkdir(parents=True, exist_ok=True)
|
21 |
+
|
22 |
+
# Ініціалізуємо S3 клієнт та завантажуємо файли якщо потрібно
|
23 |
+
if not LOCAL_DIR.exists() or not any(LOCAL_DIR.iterdir()):
|
24 |
+
print(f"Локальна директорія {LOCAL_DIR} відсутня або пуста. Починаємо завантаження...")
|
25 |
+
s3_client = initialize_s3_client()
|
26 |
+
download_s3_folder(s3_client, BUCKET_NAME, PREFIX_RETRIEVER, LOCAL_DIR)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
else:
|
28 |
+
print(f"Локальна директорія {LOCAL_DIR} вже існує і містить файли. Завантаження пропущено.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
+
# Ініціалізуємо компоненти
|
31 |
+
if initialize_components():
|
32 |
+
print("Components initialized successfully!")
|
33 |
+
app = create_gradio_interface()
|
34 |
+
app.launch(share=True)
|
35 |
+
else:
|
36 |
+
print("Failed to initialize components. Please check the paths and try again.", file=sys.stderr)
|
37 |
+
sys.exit(1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
+
except Exception as e:
|
40 |
+
print(f"Critical error during startup: {str(e)}", file=sys.stderr)
|
41 |
+
sys.exit(1)
|
|
|
|
|
|
|
|
|
|
search.py
ADDED
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Tuple, List, Optional
|
2 |
+
from llama_index.core.schema import NodeWithScore
|
3 |
+
import sys
|
4 |
+
from initialize import app_state
|
5 |
+
from utils import get_links_html, get_links_html_lp
|
6 |
+
|
7 |
+
|
8 |
+
async def search_with_ai_action(legal_position_json: dict) -> Tuple[str, Optional[List[NodeWithScore]]]:
|
9 |
+
try:
|
10 |
+
if app_state.retriever_bm25 is None:
|
11 |
+
raise ValueError("Retriever is not initialized")
|
12 |
+
|
13 |
+
query_text = (
|
14 |
+
f"{legal_position_json['title']}: "
|
15 |
+
f"{legal_position_json['text']}: "
|
16 |
+
f"{legal_position_json['proceeding']}: "
|
17 |
+
f"{legal_position_json['category']}"
|
18 |
+
)
|
19 |
+
|
20 |
+
nodes = await app_state.retriever_bm25.aretrieve(query_text)
|
21 |
+
|
22 |
+
sources_output = "\n **Результати пошуку (наявні правові позиції ВСУ):** \n\n"
|
23 |
+
for index, node in enumerate(nodes, start=1):
|
24 |
+
source_title = node.node.metadata.get('title')
|
25 |
+
doc_ids = node.node.metadata.get('doc_id')
|
26 |
+
lp_ids = node.node.metadata.get('lp_id')
|
27 |
+
links = get_links_html(doc_ids)
|
28 |
+
links_lp = get_links_html_lp(lp_ids)
|
29 |
+
sources_output += f"\n[{index}] *{source_title}* {links_lp} 👉 Score: {node.score} {links}\n"
|
30 |
+
|
31 |
+
return sources_output, nodes
|
32 |
+
|
33 |
+
except Exception as e:
|
34 |
+
error_message = f"Error during search: {str(e)}"
|
35 |
+
print(error_message, file=sys.stderr)
|
36 |
+
return error_message, None
|
utils.py
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import requests
|
2 |
+
from bs4 import BeautifulSoup
|
3 |
+
from typing import List, Union
|
4 |
+
|
5 |
+
def parse_doc_ids(doc_ids: Union[List[str], str, None]) -> List[str]:
|
6 |
+
if doc_ids is None:
|
7 |
+
return []
|
8 |
+
if isinstance(doc_ids, list):
|
9 |
+
return [str(id).strip('[]') for id in doc_ids]
|
10 |
+
if isinstance(doc_ids, str):
|
11 |
+
cleaned = doc_ids.strip('[]').replace(' ', '')
|
12 |
+
if cleaned:
|
13 |
+
return [id.strip() for id in cleaned.split(',')]
|
14 |
+
return []
|
15 |
+
|
16 |
+
def get_links_html(doc_ids: Union[List[str], str, None]) -> str:
|
17 |
+
parsed_ids = parse_doc_ids(doc_ids)
|
18 |
+
if not parsed_ids:
|
19 |
+
return ""
|
20 |
+
links = [f"[Рішення ВС: {doc_id}](https://reyestr.court.gov.ua/Review/{doc_id})"
|
21 |
+
for doc_id in parsed_ids]
|
22 |
+
return ", ".join(links)
|
23 |
+
|
24 |
+
def parse_lp_ids(lp_ids: Union[str, int, None]) -> List[str]:
|
25 |
+
if lp_ids is None:
|
26 |
+
return []
|
27 |
+
if isinstance(lp_ids, (str, int)):
|
28 |
+
cleaned = str(lp_ids).strip('[]').replace(' ', '')
|
29 |
+
if cleaned:
|
30 |
+
return [cleaned]
|
31 |
+
return []
|
32 |
+
|
33 |
+
def get_links_html_lp(lp_ids: Union[str, int, None]) -> str:
|
34 |
+
parsed_ids = parse_lp_ids(lp_ids)
|
35 |
+
if not parsed_ids:
|
36 |
+
return ""
|
37 |
+
links = [f"[ПП ВС: {lp_id}](https://lpd.court.gov.ua/home/search/{lp_id})"
|
38 |
+
for lp_id in parsed_ids]
|
39 |
+
return ", ".join(links)
|
40 |
+
|
41 |
+
def extract_court_decision_text(url: str) -> str:
|
42 |
+
response = requests.get(url)
|
43 |
+
soup = BeautifulSoup(response.content, 'html.parser')
|
44 |
+
|
45 |
+
unwanted_texts = [
|
46 |
+
"Доступ до Реєстру здійснюється в тестовому (обмеженому) режимі.",
|
47 |
+
"З метою упередження перешкоджанню стабільній роботі Реєстру"
|
48 |
+
]
|
49 |
+
|
50 |
+
decision_text = ""
|
51 |
+
for paragraph in soup.find_all('p'):
|
52 |
+
text = paragraph.get_text(separator="\n").strip()
|
53 |
+
if not any(unwanted_text in text for unwanted_text in unwanted_texts):
|
54 |
+
decision_text += text + "\n"
|
55 |
+
return decision_text.strip()
|