NCTCMumbai's picture
Update app.py
66dc0a2 verified
raw
history blame
18.9 kB
from ragatouille import RAGPretrainedModel
import subprocess
import json
import spaces
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 huggingface_hub import InferenceClient
from os import getenv
from backend.query_llm import generate_hf, generate_openai
from backend.semantic_search import table, retriever
from huggingface_hub import InferenceClient
VECTOR_COLUMN_NAME = "vector"
TEXT_COLUMN_NAME = "text"
HF_TOKEN = getenv("HUGGING_FACE_HUB_TOKEN")
proj_dir = Path(__file__).parent
# Setting up the logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1",token=HF_TOKEN)
# 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
# firebase_admin.initialize_app(cred)
# # # Create a reference to the Firestore database
# db = firestore.client()
# #db usage
# collection_name = 'Nirvachana' # Replace with your collection name
# field_name = 'message_count' # Replace with your field name for count
# Examples
examples = ['My transhipment cargo is missing','can u explain and tabulate difference between b 17 bond and a warehousing bond',
'What are benefits of the AEO Scheme and eligibility criteria?',
'What are penalties for customs offences? ', 'what are penalties to customs officers misusing their powers under customs act?','What are eligibility criteria for exemption from cost recovery charges','list in detail what is procedure for obtaining new approval for openeing a CFS attached to an ICD']
# 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 = 25
top_k_rank = 20
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/cbseclass10index')
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
yield history, prompt_html
print('Final history is ',history)
#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
yield history, prompt_html
print('Final history is ',history)
#store_message(db,history[-1][0],history[-1][1],cross_encoder)
def system_instructions(question_difficulty, topic,documents_str):
return f"""<s> [INST] Your are a great teacher and your task is to create 10 questions with 4 choices with a {question_difficulty} difficulty about topic request " {topic} " only from the below given documents, {documents_str} then create an answers. Index in JSON format, the questions as "Q#":"" to "Q#":"", the four choices as "Q#:C1":"" to "Q#:C4":"", and the answers as "A#":"Q#:C#" to "A#":"Q#:C#". [/INST]"""
#with gr.Blocks(theme='Insuz/SimpleIndigo') as demo:
with gr.Blocks(theme='NoCrypt/miku') as CHATBOT:
with gr.Row():
with gr.Column(scale=10):
# gr.Markdown(
# """
# # Theme preview: `paris`
# To use this theme, set `theme='earneleh/paris'` in `gr.Blocks()` or `gr.Interface()`.
# You can append an `@` and a semantic version expression, e.g. @>=1.0.0,<2.0.0 to pin to a given version
# of this theme.
# """
# )
gr.HTML(value="""<div style="color: #FF4500;"><h1>ADWITIYA-</h1> <h1><span style="color: #008000">Custom Manual Chatbot and Quizbot</span></h1>
</div>""", elem_id='heading')
gr.HTML(value=f"""
<p style="font-family: sans-serif; font-size: 16px;">
Using GenAI for CBIC Capacity Building - A free chat bot developed by National Customs Targeting Center using Open source LLMs for CBIC Officers
</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: 14px;">Developed by NCTC,Mumbai . Suggestions may be sent to <a href="mailto:nctc-admin@gov.in" style="color: #00008B; font-style: italic;">ramyadevi1607@yahoo.com</a>.</p>""", elem_id='Sub-heading1 ')
with gr.Column(scale=3):
gr.Image(value='logo.png',height=200,width=200)
# gr.HTML(value="""<div style="color: #FF4500;"><h1>CHEERFULL CBSE-</h1> <h1><span style="color: #008000">AI Assisted Fun Learning</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 Artificial Intelligence Chatbot assistant trained on CBSE Class 10 Science Notes to engage and help students and teachers of Puducherry.
# </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 K M Ramyasri , PGT . Suggestions may be sent to <a href="mailto:ramyadevi1607@yahoo.com" style="color: #00008B; font-style: italic;">ramyadevi1607@yahoo.com</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)
RAG_db=gr.State()
with gr.Blocks(title="Quiz Maker", theme=gr.themes.Default(primary_hue="green", secondary_hue="green"), css="style.css") as QUIZBOT:
def load_model():
RAG= RAGPretrainedModel.from_pretrained("colbert-ir/colbertv2.0")
RAG_db.value=RAG.from_index('.ragatouille/colbert/indexes/cbseclass10index')
return 'Ready to Go!!'
with gr.Column(scale=4):
gr.HTML("""
<center>
<h1><span style="color: purple;">ADWITIYA</span> Customs Manual Quizbot</h1>
<h2>Generative AI-powered Capacity building for Training Officers</h2>
<i>⚠️ NACIN Faculties create quiz from any topic dynamically for classroom evaluation after their sessions ! ⚠️</i>
</center>
""")
#gr.Warning('Retrieving using ColBERT.. First time query will take a minute for model to load..pls wait')
with gr.Column(scale=2):
load_btn = gr.Button("Click to Load!πŸš€")
load_text=gr.Textbox()
load_btn.click(load_model,[],load_text)
topic = gr.Textbox(label="Enter the Topic for Quiz", placeholder="Write any topic/details from Customs Manual")
with gr.Row():
radio = gr.Radio(
["easy", "average", "hard"], label="How difficult should the quiz be?"
)
generate_quiz_btn = gr.Button("Generate Quiz!πŸš€")
quiz_msg=gr.Textbox()
question_radios = [gr.Radio(visible=False), gr.Radio(visible=False), gr.Radio(
visible=False), gr.Radio(visible=False), gr.Radio(visible=False), gr.Radio(visible=False), gr.Radio(visible=False), gr.Radio(
visible=False), gr.Radio(visible=False), gr.Radio(visible=False)]
print(question_radios)
@spaces.GPU
@generate_quiz_btn.click(inputs=[radio, topic], outputs=[quiz_msg]+question_radios, api_name="generate_quiz")
def generate_quiz(question_difficulty, topic):
top_k_rank=10
RAG_db_=RAG_db.value
documents_full=RAG_db_.search(topic,k=top_k_rank)
generate_kwargs = dict(
temperature=0.2,
max_new_tokens=4000,
top_p=0.95,
repetition_penalty=1.0,
do_sample=True,
seed=42,
)
question_radio_list = []
count=0
while count<=3:
try:
documents=[item['content'] for item in documents_full]
document_summaries = [f"[DOCUMENT {i+1}]: {summary}{count}" for i, summary in enumerate(documents)]
documents_str='\n'.join(document_summaries)
formatted_prompt = system_instructions(
question_difficulty, topic,documents_str)
print(formatted_prompt)
pre_prompt = [
{"role": "system", "content": formatted_prompt}
]
response = client.text_generation(
formatted_prompt, **generate_kwargs, stream=False, details=False, return_full_text=False,
)
output_json = json.loads(f"{response}")
print(response)
print('output json', output_json)
global quiz_data
quiz_data = output_json
for question_num in range(1, 11):
question_key = f"Q{question_num}"
answer_key = f"A{question_num}"
question = quiz_data.get(question_key)
answer = quiz_data.get(quiz_data.get(answer_key))
if not question or not answer:
continue
choice_keys = [f"{question_key}:C{i}" for i in range(1, 5)]
choice_list = []
for choice_key in choice_keys:
choice = quiz_data.get(choice_key, "Choice not found")
choice_list.append(f"{choice}")
radio = gr.Radio(choices=choice_list, label=question,
visible=True, interactive=True)
question_radio_list.append(radio)
if len(question_radio_list)==10:
break
else:
print('10 questions not generated . So trying again!')
count+=1
continue
except Exception as e:
count+=1
print(f"Exception occurred: {e}")
if count==3:
print('Retry exhausted')
gr.Warning('Sorry. Pls try with another topic !')
else:
print(f"Trying again..{count} time...please wait")
continue
print('Question radio list ' , question_radio_list)
return ['Quiz Generated!']+ question_radio_list
check_button = gr.Button("Check Score")
score_textbox = gr.Markdown()
@check_button.click(inputs=question_radios, outputs=score_textbox)
def compare_answers(*user_answers):
user_anwser_list = []
user_anwser_list = user_answers
answers_list = []
for question_num in range(1, 20):
answer_key = f"A{question_num}"
answer = quiz_data.get(quiz_data.get(answer_key))
if not answer:
break
answers_list.append(answer)
score = 0
for item in user_anwser_list:
if item in answers_list:
score += 1
if score>5:
message = f"### Good ! You got {score} over 10!"
elif score>7:
message = f"### Excellent ! You got {score} over 10!"
else:
message = f"### You got {score} over 10! Dont worry . You can prepare well and try better next time !"
return message
demo = gr.TabbedInterface([CHATBOT,QUIZBOT], ["AI ChatBot", "AI Quizbot"])
demo.queue()
demo.launch(debug=True)