File size: 5,190 Bytes
54d04d4 7cfa59a 54d04d4 7cfa59a 54d04d4 7cfa59a 54d04d4 7cfa59a 54d04d4 7cfa59a |
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 |
""" Interactive RAG chatbot using Gradio.
Name: Constitution Of India RAG Chatbot
Phi3-mini-4k + MiniLM + chromaDB
"""
import os
from pathlib import Path
from turtle import undo
import gradio as gr
from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, PromptTemplate,Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.huggingface import HuggingFaceLLM
from llama_index.core.memory import ChatMemoryBuffer
from textblob import TextBlob
from typing import List, Tuple
from transformers import AutoTokenizer
EMBED_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
CHROMA_DB_PATH = "./chroma_db"
MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"
MODEL_NAME = "Qwen/Qwen2-1.5B-Instruct" # ✅ No tokenizer bugs
MAX_HISTORY_TOKENS = 8000
TOP_K = 4
# user query embeded with this model
Settings.embed_model = HuggingFaceEmbedding(model_name=EMBED_MODEL, device="cpu")
# phi3 LLm (downloads ~2GB on first use)
# Model name and its tokenizer name are the same most of the times. check HF for tokenizer name if not found.
llm = HuggingFaceLLM(
model_name=MODEL_NAME,
tokenizer_name=MODEL_NAME,
context_window=32768,
max_new_tokens=512,
device_map="cpu")
qa_prompt = PromptTemplate(
"""<|im_start|>system
You are an expert on the Constitution of India. Your job is to answer questions using ONLY the provided Constitution text excerpts.
RULES (MANDATORY):
1. Answer using ONLY the context provided below
2. If the answer is NOT in the context, respond EXACTLY: "❌ Not found in Constitution of India"
3. Cite specific Article/Section numbers when possible
4. Never use external knowledge or general facts
5. Be precise, legal, and constitutional in tone
6. If asked about creator - Respond with Abhijeet M
CONTEXT FROM CONSTITUTION:
{context_str}
QUESTION: {query_str}<|im_end|>
<|im_start|>assistant"""
)
class ConstitutionRAGChatBot:
def __init__(self):
if not os.path.exists(CHROMA_DB_PATH):
raise FileNotFoundError(f"ChromaDB index not found at {CHROMA_DB_PATH}. Run indexing first.")
# load index from storage -- already computed by index_builder.py
storage_context = StorageContext.from_defaults(persist_dir=CHROMA_DB_PATH)
self.index = load_index_from_storage(storage_context)
self.query_engine = self.index.as_query_engine(llm=llm, chat_mode=True, similarity_top_k=TOP_K, response_mode="compact", text_qa_template=qa_prompt, memory=ChatMemoryBuffer.from_defaults(token_limit=MAX_HISTORY_TOKENS))
# self.chat_engine = self.index.as_chat_engine(
# chat_mode="context",
# query_engine=self.query_engine, # Injects your custom prompt + settings
# memory=ChatMemoryBuffer.from_defaults(token_limit=MAX_HISTORY_TOKENS))
def preprocess_query(self, query: str) -> str:
""" Preprocess user query to improve accuracy. """
# correct spelling
corrected_query = TextBlob(query.strip()).correct()
return str(corrected_query)
def chat(self, message: str, history: List[Tuple[str, str]]) -> str:
""" Callback """
if not message.strip():
return "Please, Stick to the questions regarding the Constitutions. Thanks!"
# for user_msg, bot_msg in history[-3:]: # Last 3 exchanges
# print ('History:\n')
# print(user_msg,"\n", bot_msg)
# print ('Ends..:\n')
try:
clean_query = self.preprocess_query(message)
# query RAG (auto embed, retrives, generate)
response = self.query_engine.query(clean_query)
# response = self.chat_engine.chat(clean_query)
# print(f"📜 Retrieved context: {response.get_formatted_sources()}")
if "Not Found" in response.response.lower():
return "Its my Bad. Might be there is no information on this topic into the constitution of India or Legal language is too hard for me too.. ;)"
return response.response
except Exception as e:
return f"Error: {str(e)}.\n Try rephrasing your question in less workds"
def create_demo():
rag = ConstitutionRAGChatBot()
demo = gr.ChatInterface(
fn=rag.chat,
title = 'YourHonor',
description="Ask precise questions about Articles, Rights, Duties, Amendments. ",
theme="soft",
examples=[
"What does Article 14 say?",
"Fundamental Rights list?",
"President election process?",
"Emergency provisions?",
],
cache_examples=False,
retry_btn="Ask Again",
undo_btn="Undo",
submit_btn="Order!Order!"
)
return demo
if __name__ == "__main__":
# Local test
demo = create_demo()
# demo.launch(server_name="0.0.0.0", server_port=7860, debug=True)
# demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
demo.launch() |