File size: 3,284 Bytes
5b83b9f
 
 
e1c4a41
5b83b9f
74d0b2a
 
 
5b83b9f
e1c4a41
 
 
 
 
 
 
 
 
 
 
 
 
 
5b83b9f
e1c4a41
 
5b83b9f
e1c4a41
74d0b2a
5b83b9f
 
 
e1c4a41
5b83b9f
74d0b2a
 
e1c4a41
 
 
 
74d0b2a
e1c4a41
5b83b9f
 
74d0b2a
5b83b9f
e1c4a41
 
 
5b83b9f
 
e1c4a41
 
 
 
 
 
 
 
5b83b9f
 
74d0b2a
5b83b9f
74d0b2a
 
 
 
 
8162b2b
74d0b2a
 
 
 
 
 
8162b2b
74d0b2a
 
 
 
 
 
 
 
 
 
5b83b9f
74d0b2a
 
 
 
 
e1c4a41
74d0b2a
 
 
 
e1c4a41
 
5b83b9f
74d0b2a
e1c4a41
74d0b2a
e1c4a41
 
74d0b2a
 
 
 
e1c4a41
 
5b83b9f
 
e1c4a41
3815111
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
import os
import torch
import logging
from dotenv import load_dotenv

from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.vectorstores import Chroma
from langchain_huggingface import (
    HuggingFaceEndpoint,
    HuggingFaceEmbeddings,
    ChatHuggingFace,
)

# Load environment variables
load_dotenv()

# Logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Device
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# Globals
retrieval_chain = None
chat_history = []
embeddings = None


def init_llm():
    global embeddings
    logger.info("Initializing embeddings...")
    embeddings = HuggingFaceEmbeddings(
        model_name="sentence-transformers/all-MiniLM-L6-v2",
        model_kwargs={"device": DEVICE},
    )
    logger.info("Embeddings initialized successfully.")


def process_document(document_path):
    global retrieval_chain

    logger.info(f"Processing document: {document_path}")

    loader = PyPDFLoader(document_path)
    documents = loader.load()

    if not documents:
        raise ValueError("No text found in PDF (possibly scanned PDF).")

    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=1024,
        chunk_overlap=64,
    )

    texts = text_splitter.split_documents(documents)
    db = Chroma.from_documents(texts, embedding=embeddings)
    retriever = db.as_retriever(search_kwargs={"k": 4})

    # Hugging Face API token
    hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
    if not hf_token:
        raise ValueError("HUGGINGFACEHUB_API_TOKEN not found in environment variables")

    # Base LLM forcing the standard 'hf-inference' provider
    base_llm = HuggingFaceEndpoint(
        repo_id="meta-llama/Llama-3.1-8B-Instruct",
        task="text-generation",
        huggingfacehub_api_token=hf_token,
        temperature=0.1,
        max_new_tokens=600,
        provider="hf-inference",
    )

    chat_model = ChatHuggingFace(llm=base_llm)

    system_prompt = (
        "You are an AI assistant for document analysis. "
        "Use the following pieces of retrieved context to answer "
        "the user's question accurately. If you don't know the answer, "
        "say that you don't know.\n\n"
        "{context}"
    )
    
    prompt = ChatPromptTemplate.from_messages([
        ("system", system_prompt),
        ("human", "{input}"),
    ])

    question_answer_chain = create_stuff_documents_chain(chat_model, prompt)
    retrieval_chain = create_retrieval_chain(retriever, question_answer_chain)

    logger.info("Document processed and LCEL retrieval chain ready.")


def process_prompt(prompt):
    global chat_history, retrieval_chain

    if retrieval_chain is None:
        return "Please upload a PDF first."

    # Invoke modern LCEL chain
    response = retrieval_chain.invoke({"input": prompt})
    answer = response.get("answer", "I couldn't generate a response.")
    
    chat_history.append((prompt, answer))
    return answer.strip()


# Initialize on startup
init_llm()