File size: 10,850 Bytes
5887a43
 
 
 
5103cb0
5887a43
 
 
5103cb0
5887a43
 
5103cb0
5887a43
 
 
 
5103cb0
5887a43
 
 
5103cb0
5887a43
5103cb0
 
5887a43
5103cb0
 
 
5887a43
5103cb0
 
 
 
 
5887a43
5103cb0
 
5887a43
5103cb0
 
 
5887a43
 
5103cb0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5887a43
 
5103cb0
5887a43
 
 
 
 
 
 
5103cb0
 
5887a43
 
5103cb0
 
 
5887a43
 
5103cb0
 
5887a43
5103cb0
5887a43
5103cb0
 
 
5887a43
5103cb0
 
5887a43
5103cb0
 
5887a43
 
 
 
 
 
5103cb0
 
 
 
5887a43
5103cb0
 
 
 
 
 
 
 
 
 
 
 
5887a43
 
 
 
 
5103cb0
5887a43
 
 
 
 
 
 
 
 
 
 
 
5103cb0
d77386f
5887a43
 
 
5103cb0
5887a43
 
 
 
 
 
 
5103cb0
5887a43
 
 
 
 
 
 
 
 
5103cb0
5887a43
5103cb0
5887a43
5103cb0
 
 
5887a43
5103cb0
 
 
 
5887a43
5103cb0
5887a43
 
5103cb0
 
 
 
 
5887a43
 
5103cb0
 
 
 
 
5887a43
5103cb0
 
 
 
 
52d303b
5103cb0
5887a43
5103cb0
 
5887a43
5103cb0
 
 
 
 
5887a43
 
5103cb0
 
 
bfc0917
5103cb0
 
5887a43
 
 
5103cb0
 
5887a43
 
 
 
 
 
 
 
 
 
 
 
 
5103cb0
 
5887a43
5103cb0
5887a43
5103cb0
 
 
 
 
5887a43
5103cb0
5887a43
 
 
 
 
 
5103cb0
 
 
5887a43
 
 
 
 
5103cb0
5887a43
 
5103cb0
 
d77386f
 
5103cb0
d77386f
5103cb0
 
 
 
 
 
 
 
 
 
 
d77386f
5887a43
5103cb0
 
5887a43
5103cb0
 
5887a43
5103cb0
 
5887a43
 
 
 
 
 
 
 
 
 
 
 
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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
import subprocess

import streamlit as st
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma, FAISS
from langchain.embeddings import FastEmbedEmbeddings  # General embeddings from HuggingFace models.
from langchain.memory import ConversationBufferMemory
from langchain.callbacks.manager import CallbackManager
from langchain.callbacks import StreamlitCallbackHandler
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from htmlTemplates import css, bot_template, user_template
from langchain.llms import LlamaCpp, OpenAI, GooglePalm  # For loading transformer models.
from langchain.document_loaders import PyPDFLoader, TextLoader, CSVLoader
from langchain.chains import RetrievalQA
from langchain.prompts import PromptTemplate
from langchain import hub
import tempfile 
import os
import glob
import shutil
import time

# TEXT LOADERS
def get_pdf_text(pdf_docs):
    """
    Purpose: A hypothetical loader for PDF files in Python.
    Usage: Used to extract text or other information from PDF documents.
    Load Function: A load_pdf function might be used to read and extract data from a PDF file.

    input : pdf document path
    returns : extracted text 
    """
    temp_dir = tempfile.TemporaryDirectory() 
    temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) 

    with open(temp_filepath, "wb") as f:  
        f.write(pdf_docs.getvalue()) 

    pdf_loader = PyPDFLoader(temp_filepath) 
    pdf_doc = pdf_loader.load() 
    return pdf_doc 


def get_text_file(text_docs):
    """
    """
    temp_dir = tempfile.TemporaryDirectory()
    temp_filepath = os.path.join(temp_dir.name, text_docs.name)
    
    with open(temp_filepath, "wb") as f:
        f.write(text_docs.getvalue())
        
    text_loader = TextLoader(temp_filepath)
    text_doc = text_loader.load()
    return text_doc 
    
def get_csv_file(csv_docs):
    temp_dir = tempfile.TemporaryDirectory()
    temp_filepath = os.path.join(temp_dir.name, csv_docs.name)
    
    with open(temp_filepath, "wb") as f:
        f.write(csv_docs.getvalue())
        
    csv_loader = CSVLoader(temp_filepath)
    csv_doc = csv_loader.load()
    return csv_doc


# Break the documents into chunks
def get_text_chunks(documents):
    """
    For the compute purpose we will split the document into multiple smaller chunks. 

    IMPORTANT :  If the chunks too small we will miss the context and if its too large we will have longer compute time 
    """
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size= 1000,  
        chunk_overlap=200,  
    )

    text_chunks = text_splitter.split_documents(documents)  

    return text_chunks


# Save chunks to vector store
def get_vectorstore(text_chunks):
    """
    Load our vectors into chroma DB, Googles Vector Store
    """
    vectorstore = Chroma.from_documents(documents= text_chunks, 
                                                        embedding= st.session_state.embeddings,
                                                        persist_directory= "./vectordb/")

    return vectorstore
 

# Bind the Vector DB, Large Language models and Embedding models all into one container
def get_conversation_chain(vectorstore):
    """
    This is a langchain model where we will be binding the runner to infer data from LLM
    """
    model_path = st.session_state.model
    callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])

    if st.session_state.model == "Google_PaLm" :
        llm = GooglePalm(google_api_key = "AIzaSyAIo7rw6iJPWpkpOXSJk6BnuOKNSaB5muM",  
                         max_output_tokens = 4000, 
                         callback_manager=callback_manager)

    elif st.session_state.model == "Open_AIGPT-3.5-Turbo":
        llm = OpenAI(api_key = "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", 
                     callback_manager = callback_manager, 
                     max_tokens= 4000 )

    else:
        llm = LlamaCpp(model_path= model_path, 
                    n_ctx= 4000, 
                    max_tokens= 4000,
                    f16_kv = True, 
                    callback_manager = callback_manager,
                    verbose=True)

    prompt_template = """You are a personal HR Bot assistant for answering any questions about Companies policies
    You are given a question and a set of documents.
    If the user's question requires you to provide specific information from the documents, give your answer based only on the examples provided below. DON'T generate an answer that is NOT written in the provided examples.
    If you don't find the answer to the user's question with the examples provided to you below, answer that you didn't find the answer in the documentation and propose him to rephrase his query with more details.
    Use bullet points if you have to make a list, only if necessary. Use 'DOCUMENTS' as a reference point, to understand and give a consciese output in 3 or 5 sentences. 

    QUESTION: {question}

    DOCUMENTS:
    =========
    {context}
    =========
    Finish by proposing your help for anything else.
    """

    rag_prompt_custom = PromptTemplate.from_template(prompt_template)

    # prompt = hub.pull("rlm/rag-prompt")
    prompt = hub.pull("rlm/rag-prompt-mistral")
    
    conversation_chain = RetrievalQA.from_chain_type(
        llm,
        retriever= vectorstore.as_retriever(),
        chain_type_kwargs={"prompt": prompt},
    )
    conversation_chain.callback_manager = callback_manager
    conversation_chain.memory = ConversationBufferMemory()

    return conversation_chain

# an stream lit interface to handle and save our chats
def handle_userinput():

    clear = False

    # Add clear chat button
    if st.button("Clear Chat history"):
        clear = True
        st.session_state.messages = []

    # initialise our stream  lit chat interface
    if "messages" not in st.session_state:
        st.session_state.messages = [{"role": "assistant", "content": "How can I help you?"}] 

    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])

    # Clear the cash memory
    if clear:
        st.session_state.conversation.memory.clear()
        clear = False

    if prompt := st.chat_input():

        
        with st.chat_message("user"):
            st.markdown(prompt)
        
        # add user question to chat history
        st.session_state.messages.append( {"role": "user", "content": prompt})


        with st.chat_message("assistant"):
            # set up a call back handler
            st_callback = StreamlitCallbackHandler(st.container())
            message_holder = st.empty()
            full_response = ""

            # streamlit call back manager
            st.session_state.conversation.callback_manager = st_callback
            msg = st.session_state.conversation.run(prompt)
            #st.markdown(msg)
            for chunk in msg.split():
                full_response += chunk + " "
                time.sleep(0.09)

                # add a blinking cursor to simulate typing 
                message_holder.markdown(full_response + "✏️ ")

        # Display the responce
        message_holder.info(full_response)
        
        # add responce to session state
        st.session_state.messages.append({"role": "assistant", "content": full_response}) 


# Function to apply rounded edges using CSS
def add_rounded_edges(image_path="./randstad_featuredimage.png", radius=30):
    st.markdown(
        f'<style>.rounded-img{{border-radius: {radius}px; overflow: hidden;}}</style>',
        unsafe_allow_html=True,)
    st.image(image_path, use_column_width=True, output_format='auto')


def main():

    st.set_page_config(page_title="RANDSTAD",
                       page_icon=":books:")
    st.write(css, unsafe_allow_html=True)


    if "conversation" not in st.session_state:
        st.session_state.conversation = None
    if "chat_history" not in st.session_state:
        st.session_state.chat_history = None

    st.title("πŸ’¬ Randstad HR Chatbot")
    st.subheader("πŸš€ A HR powered by Generative AI")

    # default model 
    st.session_state.model = "Google_PaLm"
    # user_question = st.text_input("Ask a question about your documents:")

    st.session_state.embeddings = FastEmbedEmbeddings( model_name= "BAAI/bge-base-en-v1.5", cache_dir="./embedding_model/")

    if len(glob.glob("./vectordb/*.sqlite3")) > 0 :
        
        vectorstore = Chroma(persist_directory="./vectordb/", embedding_function=st.session_state.embeddings)
        st.session_state.conversation = get_conversation_chain(vectorstore)
        handle_userinput()

    # side bar information
    with st.sidebar:

        # calling a 
        add_rounded_edges()

        st.subheader("Select Your Embedding Model Model")
        LLM = list( glob.glob('./models/*.gguf') )
        LLM.extend(["Open_AIGPT-3.5-Turbo", "Google_PaLm"]) 
        st.session_state.model = st.selectbox( 'Models', LLM )


        st.subheader("Your documents")
        docs = st.file_uploader(
            "Upload File (pdf,text,csv...) and click 'Process'", accept_multiple_files=True)
        
        if st.button("Process"):
            with st.spinner("Processing"):
                # get pdf text
                doc_list = []

                # using the helper function below lets load our dependencies

                # Step 1 : Load the documents 
                for file in docs:
                    print('file - type : ', file.type)
                    if file.type == 'text/plain':
                        # file is .txt
                        doc_list.extend(get_text_file(file))
                    elif file.type in ['application/octet-stream', 'application/pdf']:
                        # file is .pdf
                        doc_list.extend(get_pdf_text(file))
                    elif file.type == 'text/csv':
                        # file is .csv
                        doc_list.extend(get_csv_file(file))


                # Step 2 : Break them into Chunks
                text_chunks = get_text_chunks(doc_list)

                # Step 3 : Create Embeddings and save them to Vector DB 
                vectorstore = get_vectorstore(text_chunks)

                # Step 4 : Get our conversation chain 
                st.session_state.conversation = get_conversation_chain(vectorstore)


if __name__ == '__main__':
    command = 'CMAKE_ARGS="-DLLAMA_CUBLAS=on" FORCE_CMAKE=1 pip install llama-cpp-python --no-cache-dir'
    
    # Run the command using subprocess
    try:
        subprocess.run(command, shell=True, check=True)
        print("Command executed successfully.")
    except subprocess.CalledProcessError as e:
        print(f"Error: {e}")
    main()