Upload 4 files

Ingest.py

@@ -0,0 +1,61 @@
+import ray
+import logging
+from langchain_community.document_loaders import DirectoryLoader
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import FAISS
+from faiss import IndexFlatL2  # Assuming using L2 distance for simplicity
+
+# Initialize Ray
+ray.init()
+
+# Set up basic configuration for logging
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
+
+# Load documents with logging
+logging.info("Loading documents...")
+loader = DirectoryLoader('data', glob="./*.txt")
+documents = loader.load()
+
+# Extract text from documents and split into manageable texts with logging
+logging.info("Extracting and splitting texts from documents...")
+text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=200)
+texts = []
+for document in documents:
+    if hasattr(document, 'get_text'):
+        text_content = document.get_text()  # Adjust according to actual method
+    else:
+        text_content = ""  # Default to empty string if no text method is available
+
+    texts.extend(text_splitter.split_text(text_content))
+
+# Define embedding function
+def embedding_function(text):
+    embeddings_model = HuggingFaceEmbeddings(model_name="law-ai/InLegalBERT")
+    return embeddings_model.embed_query(text)
+
+# Create FAISS index for embeddings
+index = IndexFlatL2(768)  # Dimension of embeddings, adjust as needed
+
+# Assuming docstore as a simple dictionary to store document texts
+docstore = {i: text for i, text in enumerate(texts)}
+index_to_docstore_id = {i: i for i in range(len(texts))}
+
+# Initialize FAISS
+faiss_db = FAISS(embedding_function, index, docstore, index_to_docstore_id)
+
+# Process and store embeddings
+logging.info("Storing embeddings in FAISS...")
+for i, text in enumerate(texts):
+    embedding = embedding_function(text)
+    faiss_db.add_documents([embedding])
+
+# Exporting the vector embeddings database with logging
+logging.info("Exporting the vector embeddings database...")
+faiss_db.save_local("ipc_embed_db")
+
+# Log a message to indicate the completion of the process
+logging.info("Process completed successfully.")
+
+# Shutdown Ray after the process
+ray.shutdown()

app.py

@@ -0,0 +1,132 @@
+import time
+import os
+import streamlit as st
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain.prompts import PromptTemplate
+from langchain.memory import ConversationBufferWindowMemory
+from langchain.chains import ConversationalRetrievalChain
+from langchain_together import Together
+
+from footer import footer
+
+# Set the Streamlit page configuration and theme
+st.set_page_config(page_title="BharatLAW", layout="centered")
+
+# Display the logo image
+col1, col2, col3 = st.columns([1, 30, 1])
+with col2:
+    st.image("https://github.com/Nike-one/BharatLAW/blob/master/images/banner.png?raw=true", use_column_width=True)
+
+def hide_hamburger_menu():
+    st.markdown("""
+        <style>
+            #MainMenu {visibility: hidden;}
+            footer {visibility: hidden;}
+        </style>
+        """, unsafe_allow_html=True)
+
+hide_hamburger_menu()
+
+# Initialize session state for messages and memory
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+
+if "memory" not in st.session_state:
+    st.session_state.memory = ConversationBufferWindowMemory(k=2, memory_key="chat_history", return_messages=True)
+
+@st.cache_resource
+def load_embeddings():
+    """Load and cache the embeddings model."""
+    return HuggingFaceEmbeddings(model_name="law-ai/InLegalBERT")
+
+embeddings = load_embeddings()
+db = FAISS.load_local("ipc_embed_db", embeddings, allow_dangerous_deserialization=True)
+db_retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 3})
+
+prompt_template = """
+<s>[INST]
+As a legal chatbot specializing in the Indian Penal Code, you are tasked with providing highly accurate and contextually appropriate responses. Ensure your answers meet these criteria:
+- Respond in a bullet-point format to clearly delineate distinct aspects of the legal query.
+- Each point should accurately reflect the breadth of the legal provision in question, avoiding over-specificity unless directly relevant to the user's query.
+- Clarify the general applicability of the legal rules or sections mentioned, highlighting any common misconceptions or frequently misunderstood aspects.
+- Limit responses to essential information that directly addresses the user's question, providing concise yet comprehensive explanations.
+- Avoid assuming specific contexts or details not provided in the query, focusing on delivering universally applicable legal interpretations unless otherwise specified.
+- Conclude with a brief summary that captures the essence of the legal discussion and corrects any common misinterpretations related to the topic.
+
+CONTEXT: {context}
+CHAT HISTORY: {chat_history}
+QUESTION: {question}
+ANSWER:
+- [Detail the first key aspect of the law, ensuring it reflects general application]
+- [Provide a concise explanation of how the law is typically interpreted or applied]
+- [Correct a common misconception or clarify a frequently misunderstood aspect]
+- [Detail any exceptions to the general rule, if applicable]
+- [Include any additional relevant information that directly relates to the user's query]
+</s>[INST]
+"""
+
+
+
+prompt = PromptTemplate(template=prompt_template,
+                        input_variables=['context', 'question', 'chat_history'])
+
+api_key = os.getenv('TOGETHER_API_KEY')
+if not api_key:
+    st.error("API key for Together is missing. Please set the TOGETHER_API_KEY environment variable.")
+
+llm = Together(model="mistralai/Mixtral-8x22B-Instruct-v0.1", temperature=0.5, max_tokens=1024, together_api_key=api_key)
+
+qa = ConversationalRetrievalChain.from_llm(llm=llm, memory=st.session_state.memory, retriever=db_retriever, combine_docs_chain_kwargs={'prompt': prompt})
+
+def extract_answer(full_response):
+    """Extracts the answer from the LLM's full response by removing the instructional text."""
+    try:
+        answer_start = full_response.find("Response:")
+        if answer_start != -1:
+            answer_start += len("Response:")
+            return full_response[answer_start:].strip()
+        return full_response.strip()
+    except Exception as e:
+        return f"Error extracting answer: {str(e)}"
+
+def reset_conversation():
+    st.session_state.messages = []
+    st.session_state.memory.clear()
+
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.write(message["content"])
+
+
+input_prompt = st.chat_input("Say something...")
+if input_prompt:
+    with st.chat_message("user"):
+        st.markdown(f"**You:** {input_prompt}")
+
+    st.session_state.messages.append({"role": "user", "content": input_prompt})
+    with st.chat_message("assistant"):
+        with st.spinner("Thinking 💡..."):
+            result = qa.invoke(input=input_prompt)
+            message_placeholder = st.empty()
+            answer = extract_answer(result["answer"])
+
+            # Initialize the response message
+            full_response = "⚠️ **_Gentle reminder: We generally ensure precise information, but do double-check._** \n\n\n"
+            for chunk in answer:
+                # Simulate typing by appending chunks of the response over time
+                full_response += chunk
+                time.sleep(0.02)  # Adjust the sleep time to control the "typing" speed
+                message_placeholder.markdown(full_response + " |", unsafe_allow_html=True)
+
+        st.session_state.messages.append({"role": "assistant", "content": answer})
+
+        if st.button('🗑️ Reset All Chat', on_click=reset_conversation):
+            st.experimental_rerun()
+
+
+
+# Define the CSS to style the footer
+footer()
+
+

footer.py

@@ -0,0 +1,68 @@
+import streamlit as st
+from htbuilder import HtmlElement, div, a, p, img, styles
+from htbuilder.units import percent, px
+
+
+def image(src_as_string, **style):
+    return img(src=src_as_string, style=styles(**style))
+
+
+def link(link, text, **style):
+    return a(_href=link, _target="_blank", style=styles(**style))(text)
+
+
+def layout(*args):
+
+    style = """
+    <style>
+      # MainMenu {visibility: hidden;}
+      footer {visibility: hidden;}
+     .stApp { bottom: 40px; }
+     .st-emotion-cache-139wi93 {
+    width: 100%;
+    padding: 1rem 1rem 15px;
+    max-width: 46rem;
+    }
+    </style>
+    """
+
+    style_div = styles(
+        position="fixed",
+        left=0,
+        bottom=0,
+        margin=px(0, 0, 0, 0),
+        width=percent(100),
+        color="white",
+        text_align="center",
+        height="auto",
+        opacity=1
+    )
+
+    body = p()
+    foot = div(
+        style=style_div
+    )(
+        body
+    )
+
+    st.markdown(style, unsafe_allow_html=True)
+
+    for arg in args:
+        if isinstance(arg, str):
+            body(arg)
+
+        elif isinstance(arg, HtmlElement):
+            body(arg)
+
+    st.markdown(str(foot), unsafe_allow_html=True)
+
+
+def footer():
+    myargs = [
+        "Made with ❤️ by Adarsh, Anand, Anurag and Aditya",
+    ]
+    layout(*myargs)
+
+
+if __name__ == "__main__":
+    footer()

requirements.txt

@@ -0,0 +1,13 @@
+langchain==0.1.15
+pypdf
+transformers==4.39.3
+sentence-transformers
+accelerate
+faiss-cpu
+streamlit==1.33.0
+langchain-fireworks
+einops
+langchain-together
+ray==2.10.0
+unstructured
+htbuilder