from dotenv import load_dotenv import gradio as gr import os from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, SimpleDirectoryReader, ChatPromptTemplate, Settings from llama_index.llms.huggingface import HuggingFaceInferenceAPI from llama_index.embeddings.huggingface import HuggingFaceEmbedding from sentence_transformers import SentenceTransformer # Load environment variables load_dotenv() # Configure the Llama index settings Settings.llm = HuggingFaceInferenceAPI( model_name="meta-llama/Meta-Llama-3-8B-Instruct", tokenizer_name="meta-llama/Meta-Llama-3-8B-Instruct", context_window=3000, token=os.getenv("HF_TOKEN"), max_new_tokens=512, generate_kwargs={"temperature": 0.1}, ) Settings.embed_model = HuggingFaceEmbedding( model_name="BAAI/bge-small-en-v1.5" ) # Define the directory for persistent storage and data PERSIST_DIR = "db" PDF_DIRECTORY = 'data' # Changed to the directory containing PDFs # Ensure directories exist os.makedirs(PDF_DIRECTORY, exist_ok=True) os.makedirs(PERSIST_DIR, exist_ok=True) # Variable to store current chat conversation current_chat_history = [] def data_ingestion_from_directory(): # Use SimpleDirectoryReader on the directory containing the PDF files documents = SimpleDirectoryReader(PDF_DIRECTORY).load_data() storage_context = StorageContext.from_defaults() index = VectorStoreIndex.from_documents(documents) index.storage_context.persist(persist_dir=PERSIST_DIR) def handle_query(query): chat_text_qa_msgs = [ ( "user", """ Your name is FernAI. Your aim is to provide the best service and information about RedFerns Tech company. You should provide answers based on the conversation flow and your final aim is attract the user to connect our company services. And summarize the answer try to give the answer with less words and effectively. {context_str} Question: {query_str} """ ) ] text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs) # Load index from storage storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR) index = load_index_from_storage(storage_context) # Use chat history to enhance response context_str = "" for past_query, response in reversed(current_chat_history): if past_query.strip(): context_str += f"User asked: '{past_query}'\nBot answered: '{response}'\n" query_engine = index.as_query_engine(text_qa_template=text_qa_template, context_str=context_str) answer = query_engine.query(query) if hasattr(answer, 'response'): response = answer.response elif isinstance(answer, dict) and 'response' in answer: response = answer['response'] else: response = "Sorry, I couldn't find an answer." # Update current chat history current_chat_history.append((query, response)) return response # Example usage: Process PDF ingestion from directory print("Processing PDF ingestion from directory:", PDF_DIRECTORY) data_ingestion_from_directory() # Define the function to handle predictions def predict(message,history): response = handle_query(message) return response # Create the chat interface with a custom layout function css = ''' /* Style the chat container */ .gradio-container { display: flex; flex-direction: column; width: 450px; margin: 0 auto; padding: 20px; border: 1px solid #ddd; border-radius: 10px; background-color: #fff; box-shadow: 0 4px 8px rgba(0,0,0,0.1); position: relative; height: 600px; /* Adjust the height of the container */ } /* Style the logo */ .gradio-logo { display: flex; justify-content: center; margin-bottom: 20px; } .gradio-logo img { width: 100%; max-width: 300px; } /* Style the title */ .gradio-title { text-align: center; font-weight: bold; font-size: 24px; margin-bottom: 20px; color: #4A90E2; } /* Style the chat history */ .gradio-chat-history { flex: 1; overflow-y: auto; padding: 15px; border-bottom: 1px solid #ddd; background-color: #f9f9f9; border-radius: 5px; margin-bottom: 10px; max-height: 500px; /* Increase the height of the chat history */ } /* Style the chat messages */ .gradio-message { margin-bottom: 15px; display: flex; flex-direction: column; /* Stack messages vertically */ } .gradio-message.user .gradio-message-content { background-color: #E1FFC7; align-self: flex-end; border: 1px solid #c3e6cb; border-radius: 15px 15px 0 15px; padding: 10px; font-size: 16px; margin-bottom: 5px; max-width: 80%; } .gradio-message.bot .gradio-message-content { background-color: #fff; align-self: flex-start; border: 1px solid #ced4da; border-radius: 15px 15px 15px 0; padding: 10px; font-size: 16px; margin-bottom: 5px; max-width: 80%; } .gradio-message-content { box-shadow: 0 2px 4px rgba(0,0,0,0.1); } /* Style the footer */ .gradio-footer { display: flex; padding: 10px; border-top: 1px solid #ddd; background-color: #F8D7DA; /* Light red background color */ position: absolute; bottom: 0; width: calc(100% - 40px); /* Adjust width to match container padding */ } /* Remove Gradio footer */ footer { display: none !important; background-color: #F8D7DA; } ''' # Create a custom HTML block for the logo logo_html = ''' ''' # Create a Blocks layout with the custom HTML and ChatInterface with gr.Blocks(theme=gr.themes.Monochrome(), fill_height=True,css=css) as demo: gr.HTML(logo_html) gr.ChatInterface(predict,clear_btn=None, undo_btn=None, retry_btn=None) # Launch the interface demo.launch()