File size: 10,731 Bytes
d52cca8
4a192fa
d52cca8
52b9829
 
 
d52cca8
 
 
207595d
d52cca8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52b9829
 
 
 
 
 
 
 
d52cca8
 
d63f3fb
9cb1844
d52cca8
52b9829
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b908a63
52b9829
b908a63
 
52b9829
b908a63
52b9829
ebf26c7
 
207595d
 
6da4da3
ebf26c7
 
 
 
 
 
 
 
d52cca8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4a192fa
88685eb
1e6bb68
4a192fa
 
b544630
4a192fa
 
 
 
 
 
 
 
 
 
 
 
 
ebf26c7
d52cca8
4a192fa
 
d52cca8
4a192fa
 
 
 
 
 
 
 
 
 
 
 
88685eb
f8cf0b4
88685eb
f8cf0b4
4a192fa
f8cf0b4
4a192fa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ebf26c7
d52cca8
 
 
2db09e8
efb7284
2db09e8
7a16e1c
70d484a
 
 
88685eb
70d484a
 
044757b
52c1b59
68f9e41
52b9829
68f9e41
 
52b9829
68f9e41
 
d52cca8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f918e82
d52cca8
 
 
 
52b9829
d52cca8
 
 
 
 
 
52b9829
d52cca8
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243

from ragatouille import RAGPretrainedModel
import subprocess
import json
import firebase_admin
from firebase_admin import credentials, firestore
import logging
from pathlib import Path
from time import perf_counter
from datetime import datetime
import gradio as gr
from jinja2 import Environment, FileSystemLoader
import numpy as np
from sentence_transformers import CrossEncoder

from backend.query_llm import generate_hf, generate_openai
from backend.semantic_search import table, retriever

VECTOR_COLUMN_NAME = "vector"
TEXT_COLUMN_NAME = "text"

proj_dir = Path(__file__).parent
# Setting up the logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Set up the template environment with the templates directory
env = Environment(loader=FileSystemLoader(proj_dir / 'templates'))

# Load the templates directly from the environment
template = env.get_template('template.j2')
template_html = env.get_template('template_html.j2')
service_account_key='firebase.json'
# Create a Certificate object from the service account info
cred = credentials.Certificate(service_account_key)
# Initialize the Firebase Admin SDK
firebase_admin.initialize_app(cred)

# # Create a reference to the Firestore database
db = firestore.client()

# Examples
examples = ['what is social media and what are rules related to it for expenditure monitoring ',
            'how many reports to be submitted by Expenditure observer with annexure names ?','what is expenditure limits for parlimentary constituency and assembly constituency'
            ]
#db usage
collection_name = 'Nirvachana'  # Replace with your collection name
field_name = 'message_count'  # Replace with your field name for count


def get_and_increment_value_count(db , collection_name, field_name):
    """
    Retrieves a value count from the specified Firestore collection and field,
    increments it by 1, and updates the field with the new value."""
    collection_ref = db.collection(collection_name)
    doc_ref = collection_ref.document('count_doc')  # Assuming a dedicated document for count

    # Use a transaction to ensure consistency across reads and writes
    try:
        with db.transaction() as transaction:
            # Get the current value count (or initialize to 0 if it doesn't exist)
            current_count_doc = doc_ref.get()
            current_count_data = current_count_doc.to_dict()
            if current_count_data:
                current_count = current_count_data.get(field_name, 0)
            else:
                current_count = 0
            # Increment the count
            new_count = current_count + 1
            # Update the document with the new count
            transaction.set(doc_ref, {field_name: new_count})
            return new_count
    except Exception as e:
        print(f"Error retrieving and updating value count: {e}")
        return None  # Indicate error
        
def update_count_html():
    usage_count = get_and_increment_value_count(db ,collection_name, field_name)
    ccount_html = gr.HTML(value=f"""
    <div style="display: flex; justify-content: flex-end;">
        <span style="font-weight: bold; color: maroon; font-size: 18px;">No of Usages:</span>
        <span style="font-weight: bold; color: maroon; font-size: 18px;">{usage_count}</span>
    </div>
""")
    return count_html
    
def store_message(db,query,answer,cross_encoder):
    timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
    # Create a new document reference with a dynamic document name based on timestamp
    new_completion= db.collection('Nirvachana').document(f"chatlogs_{timestamp}")
    new_completion.set({
        'query': query,
        'answer':answer,
        'created_time': firestore.SERVER_TIMESTAMP,
        'embedding': cross_encoder,
        'title': 'Expenditure observer bot'
    })


def add_text(history, text):
    history = [] if history is None else history
    history = history + [(text, None)]
    return history, gr.Textbox(value="", interactive=False)


def bot(history, cross_encoder):
    top_rerank = 15
    top_k_rank = 10
    query = history[-1][0]

    if not query:
         gr.Warning("Please submit a non-empty string as a prompt")
         raise ValueError("Empty string was submitted")

    logger.warning('Retrieving documents...')
    
    # if COLBERT RAGATATOUILLE PROCEDURE  : 
    if cross_encoder=='(HIGH ACCURATE) ColBERT':
        gr.Warning('Retrieving using ColBERT.. First time query will take a minute for model to load..pls wait')
        RAG= RAGPretrainedModel.from_pretrained("colbert-ir/colbertv2.0")
        RAG_db=RAG.from_index('.ragatouille/colbert/indexes/mockingbird')
        documents_full=RAG_db.search(query,k=top_k_rank)
        
        documents=[item['content'] for item in documents_full]
        # Create Prompt
        prompt = template.render(documents=documents, query=query)
        prompt_html = template_html.render(documents=documents, query=query)
    
        generate_fn = generate_hf
    
        history[-1][1] = ""
        for character in generate_fn(prompt, history[:-1]):
            history[-1][1] = character
            print('Final history is ',history)
            yield history, prompt_html
        store_message(db,history[-1][0],history[-1][1],cross_encoder)
    else:
        # Retrieve documents relevant to query
        document_start = perf_counter()
    
        query_vec = retriever.encode(query)
        logger.warning(f'Finished query vec')
        doc1 = table.search(query_vec, vector_column_name=VECTOR_COLUMN_NAME).limit(top_k_rank)
    
        
    
        logger.warning(f'Finished search')
        documents = table.search(query_vec, vector_column_name=VECTOR_COLUMN_NAME).limit(top_rerank).to_list()
        documents = [doc[TEXT_COLUMN_NAME] for doc in documents]
        logger.warning(f'start cross encoder {len(documents)}')
        # Retrieve documents relevant to query
        query_doc_pair = [[query, doc] for doc in documents]
        if cross_encoder=='(FAST) MiniLM-L6v2' :
               cross_encoder1 = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2') 
        elif cross_encoder=='(ACCURATE) BGE reranker':
               cross_encoder1 = CrossEncoder('BAAI/bge-reranker-base')
        
        cross_scores = cross_encoder1.predict(query_doc_pair)
        sim_scores_argsort = list(reversed(np.argsort(cross_scores)))
        logger.warning(f'Finished cross encoder {len(documents)}')
        
        documents = [documents[idx] for idx in sim_scores_argsort[:top_k_rank]]
        logger.warning(f'num documents {len(documents)}')
    
        document_time = perf_counter() - document_start
        logger.warning(f'Finished Retrieving documents in {round(document_time, 2)} seconds...')
    
        # Create Prompt
        prompt = template.render(documents=documents, query=query)
        prompt_html = template_html.render(documents=documents, query=query)
    
        generate_fn = generate_hf
    
        history[-1][1] = ""
        for character in generate_fn(prompt, history[:-1]):
            history[-1][1] = character
            print('Final history is ',history)
            yield history, prompt_html
        store_message(db,history[-1][0],history[-1][1],cross_encoder)


with gr.Blocks(theme='Insuz/SimpleIndigo') as demo:
    gr.HTML(value="""<div style="display: flex; align-items: center; justify-content: space-between;">
    <h1 style="color: #008000">NIRVACHANA - <span style="color: #008000">Expenditure Observer AI Assistant</span></h1>
    <img src='logo.png' alt="Chatbot" width="50" height="50" />
    </div>""",elem_id='heading')
    gr.HTML(value=f"""
    <p style="font-family: sans-serif; font-size: 16px;">
      A free chat bot assistant for Expenditure Observers on Compendium on Election Expenditure Monitoring using Open source LLMs. <br>
      The bot can answer questions in natural language, taking relevant extracts from the ECI document which can be accessed <a href="https://www.eci.gov.in/eci-backend/public/api/download?url=LMAhAK6sOPBp%2FNFF0iRfXbEB1EVSLT41NNLRjYNJJP1KivrUxbfqkDatmHy12e%2Fzk1vx4ptJpQsKYHA87guoLjnPUWtHeZgKtEqs%2FyzfTTYIC0newOHHOjl1rl0u3mJBSIq%2Fi7zDsrcP74v%2FKr8UNw%3D%3D" style="color: #00008B; text-decoration: none;">CLICK HERE !</a>.
    </p>
    """, elem_id='Sub-heading')
    usage_count = get_and_increment_value_count(db,collection_name, field_name)
    gr.HTML(value=f"""<p style="font-family: Arial, sans-serif; font-size: 16px;">Developed by Ramesh M IRS (C& CE) (R-19187), Suggestions may be sent to <a href="mailto:mramesh.irs@gov.in" style="color: #00008B; font-style: italic;">mramesh.irs@gov.in</a>.</p>""", elem_id='Sub-heading1 ')
    count_html = gr.HTML(value=f"""
    <div style="display: flex; justify-content: flex-end;">
        <span style="font-weight: bold; color: maroon; font-size: 18px;">No of Usages:</span>
        <span style="font-weight: bold; color: maroon; font-size: 18px;">{usage_count}</span>
    </div>
""")
   
    chatbot = gr.Chatbot(
            [],
            elem_id="chatbot",
            avatar_images=('https://aui.atlassian.com/aui/8.8/docs/images/avatar-person.svg',
                           'https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.svg'),
            bubble_full_width=False,
            show_copy_button=True,
            show_share_button=True,
            )

    with gr.Row():
        txt = gr.Textbox(
                scale=3,
                show_label=False,
                placeholder="Enter text and press enter",
                container=False,
                )
        txt_btn = gr.Button(value="Submit text", scale=1)

    cross_encoder = gr.Radio(choices=['(FAST) MiniLM-L6v2','(ACCURATE) BGE reranker','(HIGH ACCURATE) ColBERT'], value='(ACCURATE) BGE reranker',label="Embeddings", info="Only First query to Colbert may take litte time)")

    prompt_html = gr.HTML()
    # Turn off interactivity while generating if you click
    txt_msg = txt_btn.click(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
            bot, [chatbot, cross_encoder], [chatbot, prompt_html]).then(update_count_html,[],[count_html])

    # Turn it back on
    txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)

    # Turn off interactivity while generating if you hit enter
    txt_msg = txt.submit(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
            bot, [chatbot, cross_encoder], [chatbot, prompt_html]).then(update_count_html,[],[count_html])

    # Turn it back on
    txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)

    # Examples
    gr.Examples(examples, txt)

demo.queue()
demo.launch(debug=True)