updated new version deepseek-r1

app.py

@@ -1,35 +1,137 @@
 import streamlit as st
 import os
-from components.sidebar import render_sidebar
-from components.chat_ui import display_chat
-from services.llm import initialize_qa_chain, initialize_chain
-from utils.helpers import get_file_size
+import time
+from services.llm import initialize_llm, initialize_embeddings
+from services.vector_store import create_vector_store, retrive_vector_store, generate_prompt
+from services.pdf_processing import load_and_split_pdf
+from utils.helpers import extract_thoughts, response_generator
 
-# import subprocess
-# process = subprocess.Popen("ollama serve", shell=True) 
-# print(process)
+# Custom CSS for chat styling
+CHAT_CSS = """
+    <style>
+        .user-message {
+            text-align: right;
+            background-color: #3c8ce7;
+            color: white;
+            padding: 10px;
+            border-radius: 10px;
+            margin-bottom: 10px;
+            display: inline-block;
+            width: fit-content;
+            max-width: 70%;
+            margin-left: auto;
+            box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
+        }
+        .assistant-message {
+            text-align: left;
+            background-color: #d16ba5;
+            color: white;
+            padding: 10px;
+            border-radius: 10px;
+            margin-bottom: 10px;
+            display: inline-block;
+            width: fit-content;
+            max-width: 70%;
+            margin-right: auto;
+            box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
+        }
+    </style>
+"""
 
-# App Title
-st.title("DocChatAI | Chat Using Documents")
+# Streamlit UI Setup
+st.set_page_config(page_title="DocChatAI", layout="wide")
+st.title("📄 DocChatAI | Chat Using Documents")
 
-# Sidebar - Model Selection & File Upload
-selected_model, temperature, top_p, max_tokens, uploaded_file = render_sidebar()
+# Sidebar
+st.sidebar.title("DocChatAI")
+st.sidebar.subheader("Chat using PDF Document")
+st.sidebar.write("---")
 
-mode = False
-# Check if a PDF file is uploaded
-if uploaded_file is not None:
+# Model Selection
+selected_model = st.sidebar.radio("Choose Model", ["deepseek-r1:1.5b"])
+st.sidebar.write("---")
+
+# Hyperparameters
+temperature = st.sidebar.slider("Temperature", 0.0, 1.0, 0.7, 0.1)
+top_p = st.sidebar.slider("Top-p (Nucleus Sampling)", 0.0, 1.0, 0.9, 0.05)
+max_tokens = st.sidebar.number_input("Max Tokens", 10, 2048, 256, 10)
+st.sidebar.write("---")
+
+# File Upload
+uploaded_file = st.sidebar.file_uploader("📂 Upload a PDF", type=["pdf"])
+st.sidebar.write("---")
+
+# About Section
+st.sidebar.write("📌 **About Me**")
+st.sidebar.write("👤 **Name:** Deepak Yadav")
+st.sidebar.write("💡 **Bio:** Passionate about AI and Machine Learning.")
+st.sidebar.markdown("[GitHub](https://github.com/deepak7376) | [LinkedIn](https://www.linkedin.com/in/dky7376/)")
+st.sidebar.write("---")
+
+# Initialize LLM
+llm = initialize_llm(selected_model, temperature, top_p, max_tokens)
+embeddings = initialize_embeddings()
+
+# Document Handling
+retriever = None
+if uploaded_file:
     os.makedirs("docs", exist_ok=True)
     filepath = os.path.join("docs", uploaded_file.name)
 
-    with open(filepath, "wb") as temp_file:
-        temp_file.write(uploaded_file.read())
+    with open(filepath, "wb") as f:
+        f.write(uploaded_file.read())
+
+    # Load and process PDF
+    splits = load_and_split_pdf(filepath)
+    vectorstore = create_vector_store(splits, embeddings)
+    retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 3})
+
+# Apply custom CSS
+st.markdown(CHAT_CSS, unsafe_allow_html=True)
+
+# Initialize chat history
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+
+# Display previous messages
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.markdown(message["content"])
+
+# Chat Input
+if user_input := st.chat_input("💬 Ask something..."):
+    st.session_state.messages.append({"role": "user", "content": user_input})
+
+    with st.chat_message("user"):
+        st.markdown(user_input)
+
+    # Measure response time
+    start_time = time.time()
+
+    # Generate response
+    context = retrive_vector_store(retriever, user_input) if retriever else "No context"
+    query = generate_prompt(context=context, question=user_input)
+    response = llm.invoke(query)
+
+    # Calculate response time
+    response_time = round(time.time() - start_time, 2)
+
+    # Extract thoughts and main answer
+    thinking_part, main_answer = extract_thoughts(response)
+
+    # Display AI response
+    with st.chat_message("assistant"):
+        if thinking_part:
+            with st.expander("💭 Thought Process"):
+                st.markdown(thinking_part)
 
-    with st.spinner('Please wait...'):
-        qa_chain = initialize_qa_chain(filepath, selected_model, temperature, top_p, max_tokens)
-        mode = True
-else:
+        # **Formatted Response Display**
+        formatted_response = f""" 
+        {main_answer}  
 
-    qa_chain = initialize_chain(selected_model, temperature, top_p, max_tokens)
+        ⏳ **Response Time:** {response_time} seconds  
+        """
+        st.markdown(formatted_response, unsafe_allow_html=True)
 
-# Initialize and Display Chat History
-display_chat(qa_chain, mode)
+    # Save to session history
+    st.session_state.messages.append({"role": "assistant", "content": formatted_response})

components/chat_ui.py

@@ -1,89 +0,0 @@
-import streamlit as st
-# from services.llm import process_answer
-import time
-import re
-
-# Custom CSS for chat styling
-CHAT_CSS = """
-    <style>
-        .user-message {
-            text-align: right;
-            background-color: #3c8ce7;
-            color: white;
-            padding: 10px;
-            border-radius: 10px;
-            margin-bottom: 10px;
-            display: inline-block;
-            width: fit-content;
-            max-width: 70%;
-            margin-left: auto;
-            box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
-        }
-        .assistant-message {
-            text-align: left;
-            background-color: #d16ba5;
-            color: white;
-            padding: 10px;
-            border-radius: 10px;
-            margin-bottom: 10px;
-            display: inline-block;
-            width: fit-content;
-            max-width: 70%;
-            margin-right: auto;
-            box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
-        }
-    </style>
-"""
-def extract_thoughts(response_text):
-    """Extracts <think>...</think> content and the main answer."""
-    match = re.search(r"<think>(.*?)</think>", response_text, re.DOTALL)
-    if match:
-        thinking_part = match.group(1).strip()
-        main_answer = re.sub(r"<think>.*?</think>", "", response_text, flags=re.DOTALL).strip()
-    else:
-        thinking_part = None
-        main_answer = response_text.strip()
-    
-    return thinking_part, main_answer
-
-# Streamed response emulator
-def response_generator(response):
-    for word in response.split():
-        yield word + " "
-        time.sleep(0.05)
-
-def display_chat(qa_chain, mode):
-    st.markdown(CHAT_CSS, unsafe_allow_html=True)
-
-    if "messages" not in st.session_state:
-        st.session_state.messages = []
-
-    for message in st.session_state.messages:
-        with st.chat_message(message["role"]):
-            st.markdown(message["content"])
-
-    if prompt := st.chat_input("Ask something..."):
-        st.session_state.messages.append({"role": "user", "content": prompt})
-        with st.chat_message("user"):
-            st.markdown(prompt)
-        
-        # Get chat response
-        response = qa_chain.invoke({"input": prompt}) if mode else qa_chain.invoke({'context': prompt})
-        if not response:  # Handle empty responses
-            response = {'answer': "I don't know."}
-
-        if mode is False:
-            response = {'answer': response}
-
-        # Extract <think> part and main answer
-        thinking_part, main_answer = extract_thoughts(response['answer'])
-
-        # Display assistant response
-        with st.chat_message("assistant"):
-            if thinking_part:
-                with st.expander("💭 Thought Process"):
-                    st.markdown(thinking_part)  # Hidden by default, expandable
-                
-            response = st.write_stream(response_generator(main_answer))
-        
-        st.session_state.messages.append({"role": "assistant", "content": response})

components/sidebar.py

@@ -1,35 +0,0 @@
-import streamlit as st
-
-def render_sidebar():
-    st.sidebar.title("DocChatAI")
-    st.sidebar.subheader("Chat using PDF Document")
-    st.sidebar.write("-----------")
-    
-    # Model Selection
-    model_options = ["deepseek-r1:1.5b"]
-    selected_model = st.sidebar.radio("Choose Model", model_options)
-    
-    st.sidebar.write("-----------")
-    
-    # Hyperparameters
-    temperature = st.sidebar.slider("Temperature", min_value=0.0, max_value=1.0, value=0.7, step=0.1)
-    top_p = st.sidebar.slider("Top-p (Nucleus Sampling)", min_value=0.0, max_value=1.0, value=0.9, step=0.05)
-    max_tokens = st.sidebar.number_input("Max Tokens", min_value=10, max_value=2048, value=256, step=10)
-    
-    st.sidebar.write("-----------")
-    
-    # File Upload
-    uploaded_file = st.sidebar.file_uploader("Upload Documents", type=["pdf"])
-    
-    st.sidebar.write("-----------")
-    
-    # About Section
-    st.sidebar.write("About Me")
-    st.sidebar.write("Name: Deepak Yadav")
-    st.sidebar.write("Bio: Passionate about AI and machine learning.")
-    st.sidebar.write("[GitHub](https://github.com/deepak7376)")
-    st.sidebar.write("[LinkedIn](https://www.linkedin.com/in/dky7376/)")
-    
-    st.sidebar.write("-----------")
-    
-    return selected_model, temperature, top_p, max_tokens, uploaded_file

services/llm.py

@@ -1,25 +1,10 @@
-import ollama
-from langchain.chains import RetrievalQA
-from langchain.chains import create_retrieval_chain
 from langchain_ollama import OllamaLLM
-from services.pdf_processing import load_and_split_pdf
-from services.vector_store import create_vector_store
-from langchain.chains.combine_documents import create_stuff_documents_chain
-from langchain_core.prompts import ChatPromptTemplate
-from langchain.prompts import PromptTemplate
+from langchain_huggingface import HuggingFaceEmbeddings
 import streamlit as st
 
-PROMPT_TEMPLATE = """Question: {context}
-
-Answer: Let's think step by step."""
 
 @st.cache_resource
-def initialize_qa_chain(filepath, model_name, temperature, top_p, max_tokens):
-    # Load and split the PDF
-    splits = load_and_split_pdf(filepath)
-    vectordb = create_vector_store(splits)
-
-    # Use Ollama or Hugging Face LLM
+def initialize_llm(model_name, temperature, top_p, max_tokens):
     # Configure the LLM with additional parameters
     llm = OllamaLLM(
         model=model_name,
@@ -28,61 +13,9 @@ def initialize_qa_chain(filepath, model_name, temperature, top_p, max_tokens):
         max_tokens=max_tokens,   # Limit the number of tokens in the output
         top_p=top_p          # Nucleus sampling for controlling diversity
     )
-
-
-    # # Define strict retrieval-based prompting
-    # prompt_template = PromptTemplate(
-    #     template=(
-    #         "You are an AI assistant that only answers questions based on the provided document. "
-    #         "Do not use external knowledge. If you cannot find an answer in the document, respond with: 'I don't know.'\n\n"
-    #         "Document Context:\n{context}\n\n"
-    #         "User Question: {query}\n\n"
-    #         "Assistant Answer:"
-    #     ),
-    #     input_variables=["context", "query"]
-    # )
-
-    system_prompt = (
-    "Use the given context to answer the question. "
-    "If you don't know the answer, say you don't know. "
-    "Use three sentence maximum and keep the answer concise. "
-    "Context: {context}"
-)
-    prompt = ChatPromptTemplate.from_messages(
-        [
-            ("system", system_prompt),
-            ("human", "{input}"),
-        ]
-    )
-    question_answer_chain = create_stuff_documents_chain(llm, prompt)
-    chain = create_retrieval_chain(vectordb.as_retriever(), question_answer_chain)
-
-    # return RetrievalQA.from_chain_type(
-    #     llm=llm,
-    #     chain_type="stuff",
-    #     retriever=vectordb.as_retriever(),
-    #     chain_type_kwargs={"prompt": prompt_template}
-    # )
-    return chain
+    return llm
 
 @st.cache_resource
-def initialize_chain(model_name, temperature, top_p, max_tokens):
-    # Use Ollama or Hugging Face LLM
-    # Configure the LLM with additional parameters
-    llm = OllamaLLM(
-        model=model_name,
-        base_url="https://deepak7376-ollama-server.hf.space",
-        temperature=temperature,   # Controls randomness (0 = deterministic, 1 = max randomness)
-        max_tokens=max_tokens,   # Limit the number of tokens in the output
-        top_p=top_p          # Nucleus sampling for controlling diversity
-    )
-
-
-
-    prompt = ChatPromptTemplate.from_template(PROMPT_TEMPLATE)
-
-    chain = prompt | llm
-
-    return chain
-
-
+def initialize_embeddings():
+    embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+    return embeddings

services/pdf_processing.py

@@ -4,5 +4,6 @@ from langchain.text_splitter import RecursiveCharacterTextSplitter
 def load_and_split_pdf(filepath):
     loader = PyMuPDFLoader(filepath)  # Use PyMuPDFLoader instead
     documents = loader.load()
-    text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
-    return text_splitter.split_documents(documents)
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+    splits = text_splitter.split_documents(documents)
+    return splits

services/vector_store.py

@@ -1,10 +1,25 @@
 from langchain_community.vectorstores import FAISS
-# from langchain_community.embeddings import SentenceTransformerEmbeddings
-# from langchain_community.embeddings.ollama import OllamaEmbeddings
-from langchain_huggingface import HuggingFaceEmbeddings
-
-def create_vector_store(splits):
-    # embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
-    embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
-    # embeddings = OllamaEmbeddings(model="nomic-embed-text")
-    return FAISS.from_documents(splits, embeddings)
+
+def format_docs(docs):
+    return "\n\n".join(doc.page_content for doc in docs)
+
+def create_vector_store(splits, embeddings):
+    vectorstore = FAISS.from_documents(splits, embeddings)
+    return vectorstore
+
+def retrive_vector_store(retriever, query):
+    retrieved_docs = retriever.invoke(query)
+    return format_docs(retrieved_docs)
+
+def generate_prompt(context="", question=""):
+    return f""""You are DocChatAI, a helpful AI assistant built by Deepak7376.
+                If the user provides context, use it to answer the question.
+                If no context is provided, rely on general knowledge.
+                If you don't know the answer, say you don't know. 
+                Keep the answer concise.\n\n
+                "Context: <start_context> {context} </end_context>"
+
+                Human: {question}
+
+                Assistance: Let's think step by step.
+                """

utils/helpers.py

@@ -1,7 +1,27 @@
 import os
+import re
+import time
 
 def get_file_size(file):
     file.seek(0, os.SEEK_END)
     size = file.tell()
     file.seek(0)
     return size
+
+def extract_thoughts(response_text):
+    """Extracts <think>...</think> content and the main answer."""
+    match = re.search(r"<think>(.*?)</think>", response_text, re.DOTALL)
+    if match:
+        thinking_part = match.group(1).strip()
+        main_answer = re.sub(r"<think>.*?</think>", "", response_text, flags=re.DOTALL).strip()
+    else:
+        thinking_part = None
+        main_answer = response_text.strip()
+    
+    return thinking_part, main_answer
+
+# Streamed response emulator
+def response_generator(response):
+    for word in response.split():
+        yield word + " "
+        time.sleep(0.05)