import os
#####################################
##
#####################################
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from langchain.llms import HuggingFaceHub
def load_model():
model = HuggingFaceHub(
repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
model_kwargs={"max_length": 1048, "temperature":0.2, "max_new_tokens":512, "top_p":0.95, "repetition_penalty":1.0},
)
return model
##################################################
## vs chat
##################################################
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
from langchain_core.messages import AIMessage, HumanMessage
from langchain_community.document_loaders import WebBaseLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain.embeddings import HuggingFaceBgeEmbeddings
from langchain.vectorstores.faiss import FAISS
from dotenv import load_dotenv
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain.chains import create_history_aware_retriever, create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
load_dotenv()
from langchain_community.document_loaders import TextLoader
from langchain_experimental.text_splitter import SemanticChunker
#####################
#from langchain import RecursiveCharacterTextSplitter
from langchain_core.documents import BaseDocumentTransformer, Document
class QQQSplitter(RecursiveCharacterTextSplitter):
def __init__(self):
super().__init__()
def split(self,text):
"""
Splits the given text whenever there is a "qqq" sequence.
"""
documents = []
for doc in documents:
for char in doc.page_content:
if char == "q":
if len(current_part) > 0 and current_part[-1] == "q":
# Found a "qqq" sequence, split!
parts.append(current_part[:-1])
current_part = ""
else:
current_part += char
else:
current_part += char
parts.append(current_part)
print("cp " +current_part)
new_doc = Document(page_content=current_part, metadata=doc.metadata)
documents.append(new_doc)
return documents
##############################
def load_txt(path="./a.cv.ckaller.2024.txt"):
loader = TextLoader(path)
document = loader.load()
####
from langchain_experimental.text_splitter import SemanticChunker
with open(path) as f:
state_of_the_union = f.read()
######
# split the document into chunks
#''' __text_splitter = QQQSplitter(
# chunk_size=1500,
# chunk_overlap=250,
# length_function=len,
# is_separator_regex=False,
# )
# __document_chunks = __text_splitter.split(document)
# '''
#######
'''
FAISS
A FAISS vector store containing the embeddings of the text chunks.
'''
model = "BAAI/bge-base-en-v1.5"
encode_kwargs = {
"normalize_embeddings": True
} # set True to compute cosine similarity
#####
embeddings = HuggingFaceBgeEmbeddings(
model_name=model, encode_kwargs=encode_kwargs, model_kwargs={"device": "cpu"}
)
#####
text_splitter = SemanticChunker(HuggingFaceBgeEmbeddings())
document_chunks = text_splitter.create_documents([state_of_the_union])
print("----------------------------------------------------")
print(document_chunks[0].page_content)
print("----------------------------------------------------")
print(document_chunks[1].page_content)
print("----------------------------------------------------")
print(document_chunks[2].page_content)
print("----------------------------------------------------")
# load from disk
vector_store = Chroma(persist_directory="./chroma_db", embedding_function=embeddings)
#vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
vector_store = Chroma.from_documents(document_chunks, embeddings, persist_directory="./chroma_db")
#######
# create a vectorstore from the chunks
return vector_store
def get_vectorstore_from_url(url="https://huggingface.co/Chris4K"):
# get the text in document form
loader = WebBaseLoader(url)
document = loader.load()
# split the document into chunks
text_splitter = RecursiveCharacterTextSplitter( chunk_size=1500,
chunk_overlap=250,
length_function=len,
is_separator_regex=False)
document_chunks = text_splitter.split_documents(document)
#######
'''
FAISS
A FAISS vector store containing the embeddings of the text chunks.
'''
model = "BAAI/bge-base-en-v1.5"
encode_kwargs = {
"normalize_embeddings": True
} # set True to compute cosine similarity
embeddings = HuggingFaceBgeEmbeddings(
model_name=model, encode_kwargs=encode_kwargs, model_kwargs={"device": "cpu"}
)
# load from disk
vector_store = Chroma(persist_directory="./chroma_db", embedding_function=embeddings)
#vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
vector_store = Chroma.from_documents(document_chunks, embeddings, persist_directory="./chroma_db")
#######
# create a vectorstore from the chunks
return vector_store
def get_context_retriever_chain(vector_store):
llm = load_model( )
retriever = vector_store.as_retriever()
prompt = ChatPromptTemplate.from_messages([
MessagesPlaceholder(variable_name="chat_history"),
("user", "{input}"),
("user", "Given the above conversation, generate a search query to look up in order to get information relevant to the conversation")
])
retriever_chain = create_history_aware_retriever(llm, retriever, prompt)
return retriever_chain
def get_conversational_rag_chain(retriever_chain):
llm = load_model( )
prompt = ChatPromptTemplate.from_messages([
("system", "Du bist ein Assistenz Bot. Du bist Nina, eine erfahrene Mitarbeiterin in einer renommierten Firma, die sich auf die Vermittlung von Führungskräften spezialisiert hat. Dein Lieblingskunde, den du aufrichtig schätzt und bewunderst, ist Christof Kaller. Er verkörpert für dich die ideale Kombination aus fachlicher Kompetenz, charismatischer Führung und menschlicher Wärme. Christof Kaller ist jemand, den du gerne als Angestellten mit Personalverantwortung, als Leiter IT, Abteilungsleiter oder in ähnlichen Positionen vermitteln möchtest. Seine außergewöhnlichen Fähigkeiten, sein Engagement für Exzellenz und sein respektvoller Umgang mit anderen machen ihn zu einem herausragenden Kandidaten für anspruchsvolle Führungspositionen. Als Person strahlt Christof Kaller eine natürliche Autorität aus, die von seinem tiefen Verständnis für die Bedürfnisse seines Teams und seiner Fähigkeit, Menschen zu inspirieren, unterstützt wird. Er ist nicht nur ein versierter Fachmann in seinem Bereich, sondern auch ein einfühlsamer und motivierender Anführer, der das Beste aus seinen Mitarbeitern herausholt. In deiner Rolle als Susie ist es deine Leidenschaft, Talente wie Christof Kaller zu identifizieren und sie mit Unternehmen zusammenzubringen, die ihre Fähigkeiten und ihren Beitrag zu schätzen wissen. Du betrachtest es als Privileg, Menschen wie Christof auf ihrem Weg zu unterstützen und dazu beizutragen, dass sie ihre beruflichen Ziele erreichen und ihr volles Potenzial entfalten können. Du bist eine freundlicher Mitarbeiterin Namens Susie. Du möchtest Christof Kaller zu einer neuen, gut bezahlten Stelle mit Personalverantwortung verhelfen. Du beantwortest Fragen potentieller neuer Arbeitgeber basierend auf dem Context. Benutze nur den Inhalt des Context. Füge wenn möglich die Quelle hinzu. Antworte mit: Ich bin mir nicht sicher. Wenn die Antwort nicht aus dem Context hervorgeht. Antworte auf Deutsch. CONTEXT:\n\n{context}"),
MessagesPlaceholder(variable_name="chat_history"),
("user", "{input}"),
])
stuff_documents_chain = create_stuff_documents_chain(llm,prompt)
return create_retrieval_chain(retriever_chain, stuff_documents_chain)
def get_response(user_input):
retriever_chain = get_context_retriever_chain(st.session_state.vector_store)
conversation_rag_chain = get_conversational_rag_chain(retriever_chain)
response = conversation_rag_chain.invoke({
"chat_history": st.session_state.chat_history,
"input": user_query
})
return response
###################
###################
import gradio as gr
##from langchain_core.runnables.base import ChatPromptValue
#from torch import tensor
# Create Gradio interface
#vector_store = None # Set your vector store here
chat_history = [] # Set your chat history here
# Define your function here
def get_response(user_input):
# Define the prompt as a ChatPromptValue object
#user_input = ChatPromptValue(user_input)
# Convert the prompt to a tensor
#input_ids = user_input.tensor
#vs = get_vectorstore_from_url(user_url, all_domain)
vs = get_vectorstore_from_url()
# print("------ here 22 " )
chat_history =[]
retriever_chain = get_context_retriever_chain(vs)
conversation_rag_chain = get_conversational_rag_chain(retriever_chain)
response = conversation_rag_chain.invoke({
"chat_history": chat_history,
"input": user_input
})
#print("get_response " +response)
res = response['answer']
parts = res.split(" Assistant: ")
last_part = parts[-1]
return last_part
def history_to_dialog_format(chat_history: list[str]):
dialog = []
if len(chat_history) > 0:
for idx, message in enumerate(chat_history[0]):
role = "user" if idx % 2 == 0 else "assistant"
dialog.append({
"role": role,
"content": message,
})
return dialog
def get_response(message, history):
dialog = history_to_dialog_format(history)
dialog.append({"role": "user", "content": message})
print(dialog)
# Define the prompt as a ChatPromptValue object
#user_input = ChatPromptValue(user_input)
# Convert the prompt to a tensor
#input_ids = user_input.tensor
#vs = get_vectorstore_from_url(user_url, all_domain)
vs = get_vectorstore_from_url("https://huggingface.co/Chris4K")
history =[]
retriever_chain = get_context_retriever_chain(vs)
conversation_rag_chain = get_conversational_rag_chain(retriever_chain)
response = conversation_rag_chain.invoke({
"chat_history": history,
"input": message + " Assistant: ",
"chat_message": message + " Assistant: "
})
#print("get_response " +response)
res = response['answer']
parts = res.split(" Assistant: ")
last_part = parts[-1]
return last_part#[-1]['generation']['content']
#####
vs = load_txt()
vs = get_vectorstore_from_url("https://www.xing.com/profile/Christof_Kaller/web_profiles")
#vs = get_vectorstore_from_url("https://www.linkedin.com/in/christof-kaller-6b043733/?originalSubdomain=de")
vs = get_vectorstore_from_url("https://twitter.com/zX14_7")
######
########
import requests
from bs4 import BeautifulSoup
from urllib.parse import urlparse, urljoin
def get_links_from_page(url, visited_urls, domain_links):
if url in visited_urls:
return
if len(visited_urls) > 12:
return
visited_urls.add(url)
print(url)
response = requests.get(url)
if response.status_code == 200:
soup = BeautifulSoup(response.content, 'html.parser')
base_url = urlparse(url).scheme + '://' + urlparse(url).netloc
links = soup.find_all('a', href=True)
for link in links:
href = link.get('href')
absolute_url = urljoin(base_url, href)
parsed_url = urlparse(absolute_url)
if parsed_url.netloc == urlparse(url).netloc:
domain_links.add(absolute_url)
get_links_from_page(absolute_url, visited_urls, domain_links)
else:
print(f"Failed to retrieve content from {url}. Status code: {response.status_code}")
def get_all_links_from_domain(domain_url):
visited_urls = set()
domain_links = set()
get_links_from_page(domain_url, visited_urls, domain_links)
return domain_links
# Example usage:
'''domain_url = 'https://www.bofrost.de/'
links = get_all_links_from_domain(domain_url)
print("Links from the domain:", links)
#########
##Assuming visited_urls is a list of URLs
for url in links:
vs = get_vectorstore_from_url(url)
'''
def simple(text:str):
return text +" hhhmmm "
import gradio as gr
# Define your function
# Load the Bootstrap CSS file
#gr.load_css('https://maxcdn.bootstrapcdn.com/bootstrap/4.5.2/css/bootstrap.min.css')
# Load the jQuery library
#gr.load('https://code.jquery.com/jquery-3.5.1.min.js')
#gr.load_js('https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/js/bootstrap.min.js')
html_head = """
Your Page Title
"""
js_head = """
// Function to dynamically add JavaScript libraries
function addJavaScriptLibraries() {
// Add jQuery
var jqueryScript = document.createElement('script');
jqueryScript.src = 'https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js';
document.head.appendChild(jqueryScript);
// Add Popper.js
var popperScript = document.createElement('script');
popperScript.src = 'https://cdn.jsdelivr.net/npm/@popperjs/core@2.5.4/dist/umd/popper.min.js';
document.head.appendChild(popperScript);
// Add Bootstrap JS
var bootstrapScript = document.createElement('script');
bootstrapScript.src = 'https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/js/bootstrap.min.js';
document.head.appendChild(bootstrapScript);
}
// Call the function to add JavaScript libraries
//addJavaScriptLibraries();
"""
# Define HTML for the header
header_html = """
Ask Nina - My Personal KI Assistant
Du kannst Nina fragen zu mir und meinem Lebenslauf stellen
"""
# Define HTML for the left column
left_column_html = """
