Spaces:
Build error
Build error
File size: 9,645 Bytes
10672ab 7208469 10672ab 7208469 10672ab 7208469 10672ab 846d7a6 7208469 10672ab 7208469 10672ab 7208469 10672ab 846d7a6 10672ab 7208469 10672ab 846d7a6 10672ab 7208469 10672ab 7208469 10672ab 7208469 10672ab 7208469 10672ab 7208469 10672ab be906ef 10672ab be906ef 10672ab be906ef 10672ab be906ef 10672ab be906ef 10672ab be906ef 10672ab be906ef 10672ab be906ef 7208469 |
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 |
import streamlit as st
import os
import requests
import pdfplumber
import chromadb
import re
from langchain.document_loaders import PDFPlumberLoader
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_experimental.text_splitter import SemanticChunker
from langchain_chroma import Chroma
from langchain.chains import LLMChain
from langchain.prompts import PromptTemplate
from langchain_groq import ChatGroq
from prompts import rag_prompt, relevancy_prompt, relevant_context_picker_prompt, response_synth
# ----------------- Streamlit UI Setup -----------------
st.set_page_config(page_title="Blah", layout="centered")
st.title("Blah-1")
# ----------------- API Keys -----------------
os.environ["GROQ_API_KEY"] = st.secrets.get("GROQ_API_KEY", "")
# Load LLM models
llm_judge = ChatGroq(model="deepseek-r1-distill-llama-70b")
rag_llm = ChatGroq(model="mixtral-8x7b-32768")
llm_judge.verbose = True
rag_llm.verbose = True
# Clear ChromaDB cache to fix tenant issue
chromadb.api.client.SharedSystemClient.clear_system_cache()
st.title("Blah")
# ----------------- ChromaDB Persistent Directory -----------------
CHROMA_DB_DIR = "/mnt/data/chroma_db"
os.makedirs(CHROMA_DB_DIR, exist_ok=True)
# ----------------- Initialize Session State -----------------
if "pdf_loaded" not in st.session_state:
st.session_state.pdf_loaded = False
if "chunked" not in st.session_state:
st.session_state.chunked = False
if "vector_created" not in st.session_state:
st.session_state.vector_created = False
if "processed_chunks" not in st.session_state:
st.session_state.processed_chunks = None
if "vector_store" not in st.session_state:
st.session_state.vector_store = None
# ----------------- Improved Metadata Extraction -----------------
def extract_metadata(pdf_path):
"""Extracts title, author, emails, and affiliations from PDF."""
with pdfplumber.open(pdf_path) as pdf:
metadata = pdf.metadata or {}
# Extract title
title = metadata.get("Title", "").strip()
if not title and pdf.pages:
text = pdf.pages[0].extract_text()
title_match = re.search(r"(?i)title[:\-]?\s*(.*)", text or "")
title = title_match.group(1) if title_match else text.split("\n")[0] if text else "Untitled Document"
# Extract author
author = metadata.get("Author", "").strip()
if not author and pdf.pages:
author_match = re.search(r"(?i)by\s+([A-Za-z\s,]+)", pdf.pages[0].extract_text() or "")
author = author_match.group(1).strip() if author_match else "Unknown Author"
# Extract emails
emails = re.findall(r"[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}", pdf.pages[0].extract_text() or "")
email_str = ", ".join(emails) if emails else "No emails found"
# Extract affiliations
affiliations = re.findall(r"(?:Department|Faculty|Institute|University|College|School)\s+[\w\s]+", pdf.pages[0].extract_text() or "")
affiliation_str = ", ".join(affiliations) if affiliations else "No affiliations found"
return title, author, email_str, affiliation_str
# ----------------- Step 1: Choose PDF Source -----------------
pdf_source = st.radio("Upload or provide a link to a PDF:", ["Upload a PDF file", "Enter a PDF URL"], index=0, horizontal=True)
if pdf_source == "Upload a PDF file":
uploaded_file = st.file_uploader("Upload your PDF file", type=["pdf"])
if uploaded_file:
st.session_state.pdf_path = "/mnt/data/temp.pdf"
with open(st.session_state.pdf_path, "wb") as f:
f.write(uploaded_file.getbuffer())
st.session_state.pdf_loaded = False
st.session_state.chunked = False
st.session_state.vector_created = False
elif pdf_source == "Enter a PDF URL":
pdf_url = st.text_input("Enter PDF URL:")
if pdf_url and not st.session_state.pdf_loaded:
with st.spinner("🔄 Downloading PDF..."):
try:
response = requests.get(pdf_url)
if response.status_code == 200:
st.session_state.pdf_path = "/mnt/data/temp.pdf"
with open(st.session_state.pdf_path, "wb") as f:
f.write(response.content)
st.session_state.pdf_loaded = False
st.session_state.chunked = False
st.session_state.vector_created = False
st.success("✅ PDF Downloaded Successfully!")
else:
st.error("❌ Failed to download PDF. Check the URL.")
except Exception as e:
st.error(f"Error downloading PDF: {e}")
# ----------------- Process PDF -----------------
if not st.session_state.pdf_loaded and "pdf_path" in st.session_state:
with st.spinner("🔄 Processing document... Please wait."):
loader = PDFPlumberLoader(st.session_state.pdf_path)
docs = loader.load()
st.json(docs[0].metadata)
# Extract metadata
title, author, email_str, affiliation_str = extract_metadata(st.session_state.pdf_path)
# Display extracted metadata
st.subheader("📄 Extracted Document Metadata")
st.write(f"**Title:** {title}")
st.write(f"**Author:** {author}")
st.write(f"**Emails:** {email_str}")
st.write(f"**Affiliations:** {affiliation_str}")
# Embedding Model
model_name = "nomic-ai/modernbert-embed-base"
embedding_model = HuggingFaceEmbeddings(model_name=model_name, model_kwargs={"device": "cpu"}, encode_kwargs={'normalize_embeddings': False})
# Convert metadata into a retrievable chunk
metadata_text = f"Title: {title}\nAuthor: {author}\nEmails: {email_str}\nAffiliations: {affiliation_str}"
metadata_doc = {"page_content": metadata_text, "metadata": {"source": "metadata"}}
# Prevent unnecessary re-chunking
if not st.session_state.chunked:
text_splitter = SemanticChunker(embedding_model)
document_chunks = text_splitter.split_documents(docs)
document_chunks.insert(0, metadata_doc) # Insert metadata as a retrievable document
st.session_state.processed_chunks = document_chunks
st.session_state.chunked = True
st.session_state.pdf_loaded = True
st.success("✅ Document processed and chunked successfully!")
# ----------------- Setup Vector Store -----------------
if not st.session_state.vector_created and st.session_state.processed_chunks:
with st.spinner("🔄 Initializing Vector Store..."):
st.session_state.vector_store = Chroma(
persist_directory=CHROMA_DB_DIR, # <-- Ensures persistence
collection_name="deepseek_collection",
collection_metadata={"hnsw:space": "cosine"},
embedding_function=embedding_model
)
st.session_state.vector_store.add_documents(st.session_state.processed_chunks)
st.session_state.vector_created = True
st.success("✅ Vector store initialized successfully!")
# ----------------- Query Input -----------------
query = st.text_input("🔍 Ask a question about the document:")
if query:
with st.spinner("🔄 Retrieving relevant context..."):
retriever = st.session_state.vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 5})
retrieved_docs = retriever.invoke(query)
context = [d.page_content for d in retrieved_docs]
st.success("✅ Context retrieved successfully!")
# ----------------- Run Individual Chains Explicitly -----------------
context_relevancy_chain = LLMChain(llm=llm_judge, prompt=PromptTemplate(input_variables=["retriever_query", "context"], template=relevancy_prompt), output_key="relevancy_response")
relevant_context_chain = LLMChain(llm=llm_judge, prompt=PromptTemplate(input_variables=["relevancy_response"], template=relevant_context_picker_prompt), output_key="context_number")
relevant_contexts_chain = LLMChain(llm=llm_judge, prompt=PromptTemplate(input_variables=["context_number", "context"], template=response_synth), output_key="relevant_contexts")
response_chain = LLMChain(llm=rag_llm, prompt=PromptTemplate(input_variables=["query", "context"], template=rag_prompt), output_key="final_response")
response_crisis = context_relevancy_chain.invoke({"context": context, "retriever_query": query})
relevant_response = relevant_context_chain.invoke({"relevancy_response": response_crisis["relevancy_response"]})
contexts = relevant_contexts_chain.invoke({"context_number": relevant_response["context_number"], "context": context})
final_response = response_chain.invoke({"query": query, "context": contexts["relevant_contexts"]})
# ----------------- Display All Outputs -----------------
st.markdown("### Context Relevancy Evaluation")
st.json(response_crisis["relevancy_response"])
st.markdown("### Picked Relevant Contexts")
st.json(relevant_response["context_number"])
st.markdown("### Extracted Relevant Contexts")
st.json(contexts["relevant_contexts"])
st.markdown("### RAG Final Response")
st.write(final_response["final_response"])
st.subheader("context_relevancy_evaluation_chain Statement")
st.json(final_response["relevancy_response"])
st.subheader("pick_relevant_context_chain Statement")
st.json(final_response["context_number"])
st.subheader("relevant_contexts_chain Statement")
st.json(final_response["relevant_contexts"])
st.subheader("RAG Response Statement")
st.json(final_response["final_response"])
|