Create persistence_issue_persists_b1v2.py

lab/persistence_issue_persists_b1v2.py

@@ -0,0 +1,136 @@
+import os
+import chromadb
+import requests
+import streamlit as st
+from langchain.chains import LLMChain
+from langchain.prompts import PromptTemplate
+from langchain_groq import ChatGroq
+from langchain.document_loaders import PDFPlumberLoader
+from langchain_experimental.text_splitter import SemanticChunker
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_chroma import Chroma
+from prompts import rag_prompt
+
+# Set 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")
+
+# **Initialize session state variables**
+if "pdf_path" not in st.session_state:
+    st.session_state.pdf_path = None  
+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 "vector_store_path" not in st.session_state:
+    st.session_state.vector_store_path = "./chroma_langchain_db"
+if "vector_store" not in st.session_state:
+    st.session_state.vector_store = None
+if "documents" not in st.session_state:
+    st.session_state.documents = None
+
+# 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 = "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:", value="https://arxiv.org/pdf/2406.06998")
+    if pdf_url and not st.session_state.get("pdf_loaded", False):
+        with st.spinner("Downloading PDF..."):
+            try:
+                response = requests.get(pdf_url)
+                if response.status_code == 200:
+                    st.session_state.pdf_path = "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}")
+
+# Step 2: Process PDF
+if st.session_state.pdf_path and not st.session_state.get("pdf_loaded", False):
+    with st.spinner("Loading and processing PDF..."):
+        loader = PDFPlumberLoader(st.session_state.pdf_path)
+        docs = loader.load()
+        st.session_state.documents = docs
+        st.session_state.pdf_loaded = True  # ✅ Prevent re-loading
+        st.success(f"✅ **PDF Loaded!** Total Pages: {len(docs)}")
+
+# Step 3: Chunking
+if st.session_state.get("pdf_loaded", False) and not st.session_state.get("chunked", False):
+    with st.spinner("Chunking the document..."):
+        model_name = "nomic-ai/modernbert-embed-base"
+        embedding_model = HuggingFaceEmbeddings(model_name=model_name, model_kwargs={'device': 'cpu'}, encode_kwargs={'normalize_embeddings': False})
+        text_splitter = SemanticChunker(embedding_model)
+        documents = text_splitter.split_documents(st.session_state.documents)
+        st.session_state.documents = documents  # ✅ Store chunked docs
+        st.session_state.chunked = True  # ✅ Prevent re-chunking
+        st.success(f"✅ **Document Chunked!** Total Chunks: {len(documents)}")
+
+# Step 4: Setup Vectorstore
+if st.session_state.get("chunked", False) and not st.session_state.get("vector_created", False):
+    with st.spinner("Creating vector store..."):
+        model_name = "nomic-ai/modernbert-embed-base"
+        embedding_model = HuggingFaceEmbeddings(model_name=model_name, model_kwargs={'device': 'cpu'}, encode_kwargs={'normalize_embeddings': False})
+
+        vector_store = Chroma(
+            collection_name="deepseek_collection",
+            collection_metadata={"hnsw:space": "cosine"},
+            embedding_function=embedding_model,
+            persist_directory=st.session_state.vector_store_path
+        )
+        vector_store.add_documents(st.session_state.documents)
+        num_documents = len(vector_store.get()["documents"])
+        st.session_state.vector_store = vector_store
+        st.session_state.vector_created = True  # ✅ Prevent re-creating vector store
+        st.success(f"✅ **Vector Store Created!** Total documents stored: {num_documents}")
+
+# Step 5: Query Input (this should not trigger previous steps!)
+if st.session_state.get("vector_created", False) and st.session_state.get("vector_store", None):
+    query = st.text_input("🔍 Enter a Query:")
+
+    if query:
+        with st.spinner("Retrieving relevant contexts..."):
+            retriever = st.session_state.vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 5})
+            contexts = retriever.invoke(query)
+            context_texts = [doc.page_content for doc in contexts]
+
+        st.success(f"✅ **Retrieved {len(context_texts)} Contexts!**")
+        for i, text in enumerate(context_texts, 1):
+            st.write(f"**Context {i}:** {text[:500]}...")
+
+        # **Step 6: Generate Final Response**
+        with st.spinner("Generating the final answer..."):
+            final_prompt = PromptTemplate(input_variables=["query", "context"], template=rag_prompt)
+            response_chain = LLMChain(llm=rag_llm, prompt=final_prompt, output_key="final_response")
+            final_response = response_chain.invoke({"query": query, "context": context_texts})
+
+        st.subheader("🟥 RAG Final Response")
+        st.success(final_response['final_response'])