Duplicate from heikowagner/GPT-Docker

.dockerignore

@@ -0,0 +1,12 @@
+./docker/zeppelin/logs/*
+*.openaiapikey*
+*.log
+*.log.*
+*__pycache__*
+root
+*.ipynb_checkpoints*
+.vscode
+/app/mymodels
+/app/.cache
+/app/VectorStore
+*chroma-embeddings.parquet*

.gitattributes

@@ -0,0 +1,34 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,12 @@
+./docker/zeppelin/logs/*
+*.openaiapikey*
+*.log
+*.log.*
+*__pycache__*
+root
+*.ipynb_checkpoints*
+.vscode
+/app/mymodels
+/app/.cache
+/app/VectorStore
+*chroma-embeddings.parquet*

Dockerfile

@@ -0,0 +1,52 @@
+#Navigate to your user folder cd $env:USERPROFILE\AppData\Local\Docker\wsl\data
+#Enter the following command resize-vhd -Path .\ext4.vhdx -SizeBytes 300GB, after that I was able to continue building with docker-compose!
+
+FROM python:latest AS builder
+RUN apt update -y
+RUN apt install -y git git-lfs make gcc g++ libgmp-dev libmpfr-dev libmpc-dev
+RUN git lfs install
+RUN git clone https://github.com/ggerganov/llama.cpp
+RUN cd llama.cpp && make
+RUN git clone https://huggingface.co/nyanko7/LLaMA-7B
+RUN ls -la
+RUN cp -r ./LLaMA-7B ./llama.cpp/models
+RUN ls -la ./llama.cpp/models/LLaMA-7B
+# convert the 7B model to ggml FP16 format
+WORKDIR llama.cpp
+RUN python3 -m pip install -r requirements.txt
+RUN python3 convert.py ./models/LLaMA-7B
+# quantize the model to 4-bits (using q4_0 method)
+RUN mkdir ./models/7B/
+RUN ./quantize ./models/LLaMA-7B/ggml-model-f16.bin ./models/7B/ggml-model-q4_0.bin q4_0
+
+FROM tensorflow/tensorflow:latest-gpu
+WORKDIR /app
+COPY --from=builder /llama.cpp//models/7B/ ./mymodels/LLaMA-7B
+# RUN apt-get upgrade -y
+RUN apt update -y
+RUN apt install -y git git-lfs
+RUN apt install -y make wget git gcc g++ lhasa libgmp-dev libmpfr-dev libmpc-dev flex bison gettext texinfo ncurses-dev autoconf rsync
+COPY ./requirements.txt requirements.txt
+RUN pip install -r requirements.txt
+COPY ./app .
+#RUN python load_docs.py
+#RUN --mount=type=secret,id=OPENAI_API_KEY \
+#  cat /run/secrets/OPENAI_API_KEY > .openaiapikey
+RUN echo "" > .openaiapikey
+RUN mkdir /.cache
+RUN mkdir /nltk_data
+RUN mkdir /VectorStore
+RUN mkdir /app/.cache
+RUN mkdir /mymodels
+RUN ls -la
+RUN python run.py
+RUN chmod 777 /VectorStore
+RUN chmod 777 /mymodels
+RUN chmod 777 /nltk_data
+RUN chmod 777 /.cache
+RUN chmod 777 /app/.cache
+RUN chmod 777 /app/mymodels
+RUN chmod 777 /app/VectorStore/
+CMD ["streamlit", "run", "app.py", "--server.port=7860"]
+#CMD ls -la
+EXPOSE 7860

README.md

@@ -0,0 +1,11 @@
+---
+title: myRetrievalGPT
+emoji: 🔥
+colorFrom: green
+colorTo: purple
+sdk: docker
+pinned: true
+duplicated_from: heikowagner/GPT-Docker
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app/app.py

@@ -0,0 +1,65 @@
+import streamlit as st
+import load_model
+import utils as ut
+import elements as el
+import os
+import torch
+import psutil
+
+persist_directory = load_model.persist_directory
+st.title('myRetrievalGPT')
+st.header('An GPT Retrieval example brought to you by Heiko Wagner')
+
+st.markdown('*Let $\phi$ be a word embedding mapping $W$ → $\mathbb{R}^n$ where $W$ is the word space and $\mathbb{R}^n$ is an $n$-dimensional vector space then: $\phi(king)-\phi(man)+\phi(woman)=\phi(queen)$.* ')
+
+agree = st.checkbox('Load new Documents')
+if agree:
+    el.load_files()
+else:
+
+    import torch
+    torch.cuda.empty_cache()
+
+    st.write(str( torch.cuda.is_available()) + str(psutil.virtual_memory()))
+    model_type = st.selectbox(
+        'Select the Documents to be used to answer your question',
+        ('OpenAI', 'decapoda-research/llama-7b-hf (gpu+cpu)', 'llama-7b 4bit (cpu only)',) ) 
+
+    if model_type=='OpenAI':
+        if 'openai_key' not in st.session_state:
+            openai_key= st.text_area('OpenAI Key:', '')
+            if len(openai_key)>-1:
+                st.session_state['openai_key'] = openai_key
+                os.environ["OPENAI_API_KEY"] = openai_key
+        else:
+            os.environ["OPENAI_API_KEY"] = st.session_state.openai_key
+        llm= load_model.load_openai_model()
+    elif model_type=='decapoda-research/llama-7b-hf (gpu+cpu)':
+        # Add more models here
+        if not torch.cuda.is_available() and psutil.virtual_memory().available< 18254768640:
+            st.write('You do not have enough memory to use this model:' + str(psutil.virtual_memory().available))
+        else:
+            llm = load_model.load_gpu_model("decapoda-research/llama-7b-hf")
+    else:
+        llm = load_model.load_cpu_model()
+
+
+    collections = ut.retrieve_collections()
+    option = st.selectbox(
+        'Select the Documents to be used to answer your question',
+        collections )
+
+    st.write('You selected:', option['name'])
+
+    chain = load_model.create_chain(llm, collection=option['name'], model_name=option['model_name'], metadata= option['metadata'])
+    query = st.text_area('Ask a question:', 'Hallo how are you today?')
+    result = chain({"query": query + " Add a Score of the propability that your answer is correct to your answer"})
+    ut.format_result_set(result)
+
+#from langchain.chains import ConversationChain
+#from langchain.memory import ConversationBufferMemory
+
+#conversation = ConversationChain(
+#    llm=chat,
+#    memory=ConversationBufferMemory()
+#)

app/elements.py

@@ -0,0 +1,67 @@
+
+import streamlit as st
+from langchain.docstore.document import Document
+from chromadb.config import Settings
+from load_model import load_embedding
+from load_vectors import load_from_file, load_and_split, create_and_add, load_from_web
+from utils import retrieve_collections, get_chroma_client
+
+def llm_module():
+    pass
+
+def load_files():
+    
+    client = get_chroma_client()
+
+    option = st.radio(
+        "",
+        options=["Add Documents", "Start new collection"],
+    )
+
+    if option == "Add Documents":
+        collections = retrieve_collections()
+        selected_collection = st.selectbox(
+            'Add to exsisting collection or create a new one',
+            collections )
+        if st.button('Delete Collection (⚠️ This is destructive and not reversible)'):
+            client.delete_collection(name=selected_collection["name"])
+            #retrieve_collections.clear()
+            collections = retrieve_collections()
+
+        if selected_collection:
+            st.write("Selected Vectorstore:", selected_collection)
+        option = st.radio(
+            "",
+            options=["Upload Files from Local", "Upload Files from Web"],
+        )
+        if option == "Upload Files from Local":
+            st.write('Source Documents:')
+            uploaded_files = st.file_uploader("Choose a PDF file", accept_multiple_files=True)
+            chunk_size = st.text_area('chunk Size:', 1000)
+
+            if st.button('Upload'):
+                docs = load_from_file(uploaded_files)
+                sub_docs = load_and_split(docs, chunk_size=int(chunk_size))
+                vec1 = create_and_add(selected_collection["name"], sub_docs, selected_collection['model_name'], selected_collection['metadata'])
+                st.write("Upload succesful")
+        else:
+            st.write('Urls of Source Documents (Comma separated):')
+            urls = chunk_size = st.text_area('Urls:', '')
+            chunk_size = st.text_area('chunk Size:', 1000)
+            urls = urls.replace(",", "" ).replace('"', "" ).split(',')
+
+            if st.button('Upload'):
+                docs = load_from_web(urls)     
+                sub_docs = load_and_split(docs, chunk_size=int(chunk_size))
+                vec2 = create_and_add(selected_collection["name"], sub_docs, selected_collection['model_name'], selected_collection['metadata'])
+                st.write("Upload succesful")
+    else:
+        collection = st.text_area('Name of your new collection:', '')
+        model_name = st.text_area('Choose the embedding function:', "hkunlp/instructor-large")
+        if st.button('Create'):
+            if len(collection)>3:
+                ef = load_embedding(model_name)
+                metadata= {"loaded_docs":[], "Subject":"Terms Example", "model_name": ef.model_name}
+                client.create_collection(collection, embedding_function=ef, metadata=metadata) 
+                # retrieve_collections.clear()
+                st.write("Collection " +collection+" succesfully created.")

app/exploration.py

@@ -0,0 +1,51 @@
+# %%
+
+from utils import retrieve_collections, get_chroma_client
+
+
+from load_model import load_embedding
+
+#retrieve_collections()
+
+client = get_chroma_client()
+
+# %%
+client.reset()
+# %%
+collections = tuple( [collection.name for collection in client.list_collections()] )  ##Keine Embedding function in der Collection angelegt...
+
+ef = load_embedding("hkunlp/instructor-large")
+collection="heikostest2"
+client.create_collection(collection, embedding_function=ef, metadata={"loaded_docs":[]}) 
+
+
+# %%
+my_col = client.list_collections()
+
+# %%
+my_col.embedding_function
+
+# %%
+from langchain.vectorstores import Chroma
+import load_model
+
+from load_model import load_embedding
+
+persist_directory = load_model.persist_directory
+
+ef = load_embedding("hkunlp/instructor-large")
+vectorstore = Chroma(
+            collection_name="papers",
+            embedding_function=ef,
+            persist_directory=persist_directory,
+        )
+
+# %%
+query = "What did the president say about Ketanji Brown Jackson"
+docs = vectorstore.similarity_search(query)
+
+
+# %% 
+docs
+# %%
+vectorstore.similarity_search_with_score(query)

app/load_model.py

@@ -0,0 +1,122 @@
+# %%
+# git clone https://huggingface.co/nyanko7/LLaMA-7B
+# python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu117/torch2.00/index.html
+# apt-get update && apt-get install ffmpeg libsm6 libxext6  -y
+from transformers import LlamaForCausalLM, LlamaTokenizer
+from langchain.embeddings import LlamaCppEmbeddings, HuggingFaceInstructEmbeddings, OpenAIEmbeddings
+from langchain.llms import LlamaCpp, HuggingFacePipeline
+from langchain.vectorstores import Chroma
+from transformers import pipeline
+import torch
+torch.backends.cuda.matmul.allow_tf32 = True
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+import streamlit as st
+import cloudpickle
+import os
+from langchain.chains import RetrievalQA
+from langchain.indexes import VectorstoreIndexCreator
+from langchain.llms import OpenAI
+import multiprocessing
+
+from chromadb.config import Settings
+import chromadb
+
+import pathlib
+
+current_path = str( pathlib.Path(__file__).parent.resolve() )
+print(current_path)
+persist_directory = current_path + "/VectorStore"
+
+# %%
+@st.cache_resource
+def load_cpu_model():
+    """Does not work atm, bc cpu model is not persisted"""
+    model_path= "./mymodels/LLaMA-7B/ggml-model-q4_0.bin"
+    device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)}
+    llm = LlamaCpp(
+        model_path=model_path,
+        n_ctx=6000,
+        n_threads=multiprocessing.cpu_count(),
+        temperature=0.6,
+        top_p=0.95
+        )
+    
+    llama_embeddings = LlamaCppEmbeddings(model_path=model_path)
+    return llm
+
+@st.cache_resource(max_entries =1)
+def load_gpu_model(used_model):
+    torch.cuda.empty_cache()
+    tokenizer = LlamaTokenizer.from_pretrained(used_model)
+
+    if not torch.cuda.is_available():
+        device_map = {
+            "": "cpu"
+        }
+        quantization_config = BitsAndBytesConfig(load_in_8bit=True, llm_int8_enable_fp32_cpu_offload=True)
+        torch_dtype=torch.float32
+        load_in_8bit=False
+    else:
+        device_map="auto"
+        quantization_config = BitsAndBytesConfig(load_in_8bit=True, llm_int8_enable_fp32_cpu_offload=True) #atm no offload, bc device_map="auto"
+
+
+    base_model = LlamaForCausalLM.from_pretrained(
+        used_model,
+        device_map=device_map,
+        offload_folder=current_path + "/models_gpt/",
+        low_cpu_mem_usage=True,
+        quantization_config=quantization_config,
+        cache_dir = current_path + "/mymodels/"
+    )
+    pipe = pipeline(
+        "text-generation",
+        model=base_model, 
+        tokenizer=tokenizer, 
+        max_length=8000,
+        temperature=0.6,
+        top_p=0.95,
+        repetition_penalty=1.2
+    )
+    llm = HuggingFacePipeline(pipeline=pipe)
+    return llm
+
+#@st.cache_resource
+def load_openai_model(temperature=0.9):
+    return OpenAI(temperature=temperature)
+
+@st.cache_resource
+def load_openai_embedding():
+    return OpenAIEmbeddings()
+
+#@st.cache_resource
+def load_embedding(model_name):
+    embeddings = HuggingFaceInstructEmbeddings(
+            query_instruction="Represent the query for retrieval: ",
+            model_name = model_name,
+            cache_folder=current_path + "/mymodels/"
+        )
+    return embeddings
+
+def load_vectorstore(model_name, collection, metadata):
+        embeddings = load_embedding(model_name)
+        client_settings = Settings(
+            chroma_db_impl="duckdb+parquet",
+            persist_directory=persist_directory,
+            anonymized_telemetry=False
+        )
+        vectorstore = Chroma(
+            collection_name=collection,
+            embedding_function=embeddings,
+            client_settings=client_settings,
+            persist_directory=persist_directory,
+            collection_metadata=metadata
+        )
+        return vectorstore
+
+def create_chain(_llm, collection, model_name, metadata):
+    vectorstore = load_vectorstore(model_name, collection, metadata=metadata)
+    retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
+    chain = RetrievalQA.from_chain_type(llm=_llm, chain_type="stuff", retriever=retriever, return_source_documents=True)
+    return chain
+# %%

app/load_test.py

@@ -0,0 +1,29 @@
+# %%
+# %%
+import os
+import pathlib
+
+from load_model import load_embedding
+from utils import get_chroma_client
+from load_vectors import load_from_web, create_and_add, load_and_split
+
+collection="axaterms"
+client = get_chroma_client()
+# Load collection to get metadata
+loaded_collection = client.get_collection(collection)
+
+# %%
+model_name = loaded_collection.metadata['model_name']
+
+# %%
+print( loaded_collection.json() )
+
+
+# %%
+client.get_collection(collection).json()   #add documents destroys the metadata... maybe :)
+# %%
+
+#loaded_collection.modify(metadata={"Test":99})
+
+# %%
+loaded_collection.json()

app/load_vectors.py

@@ -0,0 +1,121 @@
+# %%
+import nltk
+from langchain.indexes import VectorstoreIndexCreator
+from langchain.text_splitter import CharacterTextSplitter, NLTKTextSplitter
+from langchain.document_loaders import OnlinePDFLoader, UnstructuredPDFLoader
+from langchain.vectorstores import Chroma
+from langchain.embeddings import LlamaCppEmbeddings, HuggingFaceInstructEmbeddings
+from chromadb.config import Settings
+import chromadb
+from chromadb.utils import embedding_functions
+from hashlib import sha256
+import cloudpickle
+import logging
+import os
+from load_model import load_embedding, load_vectorstore
+import torch
+import re
+import pathlib
+import tempfile
+
+
+current_path = str( pathlib.Path(__file__).parent.resolve() )
+
+os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
+nltk.download('punkt')
+
+persist_directory = current_path + "/VectorStore"
+logger = logging.getLogger()
+
+
+# %%
+
+def create_collection(collection_name, model_name, client):
+    """Not used atm"""
+    if not torch.cuda.is_available():
+        device= "cpu"
+    else:
+        device= "cuda"
+    ef = embedding_functions.InstructorEmbeddingFunction(
+        model_name=model_name, device=device)
+    client.get_or_create_collection(collection_name, embedding_function=ef)
+    return True
+
+def create_and_add(collection_name, sub_docs, model_name, metadata):
+    logging.info(f"Adding documents to {collection_name}")
+    embeddings = load_embedding(model_name)
+    vectorstore = load_vectorstore(model_name, collection_name, metadata = metadata)  
+    vectorstore.add_documents(documents=sub_docs, embedding=embeddings)
+    vectorstore.persist()
+
+    # Test Vectorstore
+    vectorstore2 = load_vectorstore(model_name, collection_name, metadata = metadata)
+    print( vectorstore2.similarity_search_with_score(query="What is a transformer llm?", k=4) )
+
+    return True
+
+def load_from_file(files):
+
+    saved_files=[]
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        for file in files:
+            temp_dir = pathlib.Path(tmpdirname)
+            file_name = os.path.join(temp_dir,file.name)
+            saved_files.append(file_name)
+            with open(file_name, mode='wb') as w:
+                w.write(file.read())
+
+        print(saved_files)
+        loaders=[UnstructuredPDFLoader(pdf) for pdf in saved_files]
+        docs = []
+        print(loaders)
+        for loader in loaders:
+            docs.extend(loader.load())
+    return docs
+
+def load_from_web(urls, cache=True):
+    docs_list = urls
+    filename=f"{current_path}/.cache/{sha256(str(urls).encode('utf-8')).hexdigest()}.pkl"
+
+    isFile = os.path.isfile(filename)
+
+    if cache and isFile:
+        logger.info("Using Cache")
+        pikd = open(filename, "rb")
+        docs = cloudpickle.load(pikd)
+    else:
+        loaders=[OnlinePDFLoader(pdf) for pdf in docs_list]
+        docs = []
+        for loader in loaders:
+            docs.extend(loader.load())
+        with open(filename, 'wb') as output:
+            cloudpickle.dump(docs, output)
+
+    #update metadata
+    i=0
+    for doc in docs:
+        doc.metadata = {'source': docs_list[i], 'url': docs_list[i], 'owner':'Heiko Wagner'}
+        i=i+1
+    return docs
+        
+def load_and_split(docs, chunk_size=700):
+    text_splitter = NLTKTextSplitter(chunk_size=chunk_size, chunk_overlap=0)
+    sub_docs = text_splitter.split_documents(docs)
+    return sub_docs
+
+def metadata_generator(doc, llm,max_token=4000):
+    #query = f"Document = {doc.page_content[1:max_token]} -> Respond a python code using a dict filling xxxx like {{'document_type': xxxx, 'summary (max. 30 letters)':'xxxx'}} resond at leat 10 letter"
+    query = f"""
+    Cluster the following Input document into topic categories based on patterns seen within the text. Also mention reasoning behind how these categories were defined.
+Output format:
+{{
+"DOCUMENT TYPE": "",
+"SUMMARY": [],
+"REASONING": ""
+}}
+
+Input document:
+{doc.page_content[1:max_token]}
+Output:
+    """
+    return llm(query)

app/playground/load_docs.py

@@ -0,0 +1,155 @@
+# %%
+from load_vectors import load_from_web, load_and_split, create_and_add
+
+docs = [
+    "https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/85ec0278-bf2f-4392-94b9-c086717fa8f6_axa_urd2022_accessible_va.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/d97a94ff-a848-474b-b802-c22afc8311cd_axa_half_year_2022_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/51954d53-c0cf-4f90-84f7-53ee27dbe4e6_axa_ri2021_va_accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/e3f52b5e-d4aa-4fc8-8bcd-f432df86e804_axa_urd_2021_en_accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/4f303cec-a12d-480b-accb-7b56f706f60e_axa-ri2020-en-accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/d6aef906-e41f-40c7-ac9c-29044e98939d_AXA_URD_2020_EN_accessible_b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F47b47783-ddd1-47c3-912f-bc6e318ebbb3_axa_half_year_2020_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ffd5a8bd8-9ef1-40eb-b953-c268c0ab4bf9_axa-ri2019-en-accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F90abd6c7-80c4-48ef-84bf-1d038670d9b7_axa-urd2019-en.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F3ef6a9cc-6215-4e58-83b5-756774ef5b73_axa_half_year_2019_financial_report2.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F0a5e0bd9-78f2-4ef8-b32c-1d3d35ddce80_axa-ri2018-en-accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F913d1869-3d11-4eb2-b013-4caedb747fab_axa-ddr2018b-en.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F476f79c9-c0c7-4ce3-88ed-4f99b3d22259_axa_half_year_2018_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F01f6966b-c26c-4935-91dc-1b296511ba8c_axa_ri2017_gb_planche.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fec440dc9-69df-41b5-a3af-5b5f4fc29670_axa_reference_document_2017c.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F72c59a61-8124-4066-a86d-bece5f41ce53_axa_us_statutory_statements_fy17.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F9237d78f-c1ac-43ca-9623-d0382a5aaaec_axa_us_statutory_statements_3q17.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ffdd639e0-2ea6-4c3f-8a42-8bca4359e858_axa_us_statutory_statements_2q17.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F877e30a9-df72-480f-ac25-edcfcd4049c2_axa_us_statutory_statements_1q17.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F6f3108fd-fabc-4dc6-a984-23eb0dca7a19_axa-ri2016-en_01.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F268bab7a-2e78-4843-844a-fd3ad2d340bc_axa_reference_document_2016.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fd2f66d05-e6ad-47a2-ab72-9bc727bd49c2_axa_half_year_2016_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F7a5f0af2-03c3-4a82-a077-46fdc52e5685_axa_us_statutory_statements_fy16.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fdd643342-e975-473d-af54-c64491252a19_axa_us_statutory_statements_3q16.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F53e10a7a-9348-40dc-935e-01fb0a1d0441_axa_us_statutory_statements_2q16.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F8906bad6-14cb-4594-b7c0-029f8fc2172d_axa_us_statutory_statements_1q16.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F2d8e525a-1161-453a-a14f-817f0f070f79_axa_activity_cr_report_2015_accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F73719a96-c3b1-456b-abaf-63b80c06968c_axa_reference_document_2015.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fe2936c1a-65f0-40db-b34b-bef9c27e91c0_axa_2015_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fdaac2a30-a3b8-4839-9331-041805836a6f_axa_us_statutory_statements_fy15.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F61a6c98a-08fb-4cb1-b6c0-4d1ef0f72aa9_axa_us_statutory_statements_3q15.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fe0689ffc-5aec-4388-a10e-26d1d1a7eb9a_axa_us_statutory_statements_2q15.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fbfa8ef5b-6533-4773-8502-5170a51735c9_axa_us_statutory_statements_1q15.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fbbb94857-f5d4-4afd-81d0-e85666883936_axa_annual+financial+report_2014.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fb826839c-76c9-48c7-b8c1-9eda7fe3b032_axa_activity_csr_report_2014_va_b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fac63e0f9-60ba-47c2-9e23-f1d25731c7ee_axa_2014_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fb7db2a55-8eb6-4131-bc03-698e4bc756d6_axa_us_statutory_statements_fy2014.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F417b48df-c585-4cb6-9d10-719d81228756_axa_us_statutory_statements_3q14.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F4586d978-6fb8-4c44-b934-e15c14143b6d_axa_us_statutory_statements_2q14.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F29cc016e-aff9-49c5-bb04-d55598aab844_axa_us_statutory_statements_1q14.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F25fee379-c187-40e7-bf3a-5fe1423cec0f_axa_annual+financial+report_2013.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F37614ed4-1fe0-483e-a0eb-0acefdedd065_axa_2013_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Febb51afc-af0e-4aff-9494-5b852b3233e5_axa_us_statutory_statements_fy2013.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fdd4cd68e-710e-4e00-ba96-c7560d738a43_axa_us_statutory_statements_3q13.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Feab93a81-859a-487c-941c-11e4ce08d5f0_axa_us_statutory_statements_2q13.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F879d09d7-8ff7-4c43-9a24-7ee44ee55404_axa_us_statutory_statements_1q13.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F9224097f-d703-4efd-8050-6553ef4336f8_axa_annual+financial+report_2012b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fcb9d1279-948a-4238-ab8f-754e9e10f2a5_axa_activity_csr_report_2012b_va.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fbed64ef2-5078-425a-a616-ffb1947e0b65_axa_2012_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fa148165a-b818-4ea1-b7ee-7949cc86ff9a_axa_us_statutory_statements_fy2012.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F601ed5e8-189d-4e59-b0d4-d1c1eedb2ffe_axa_us_statutory_statements_3q12.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F9637b674-c740-4115-9c90-3a8827516cc0_axa_us_statutory_statements_2q12.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F319a5964-ea51-4d51-96c8-cf6838047b72_axa_us_statutory_statements_1q12.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F0b75d1fe-4b11-4462-9883-4e3bc7532bf4_axa_annual+financial+report_2011.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F17e098ab-3335-4ee1-ade7-058517a952c4_axa_activity_csr_report_2011_vab.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F03996908-7e75-465e-8082-b44f02da326a_axa_us_statutory_statements_fy2011.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fea813e84-7d08-4cf2-bea1-3a01fd4bdf62_axa_us_statutory_statements_3q11.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fda28b496-275b-451d-bffd-108714eb2c39_axa_us_statutory_statements_2q11.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fd6aa2b39-896e-47cf-9882-9985c8d44276_axa_us_statutory_statements_1q11.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fc76f47f4-0917-4fb1-b1ae-78e2a4fbcef5_axa_annual+financial+report_2010c+%281%29.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F68c2771e-5ed8-41d9-bb59-f37f6403b4bf_axa_activity_csr_report_2010_vac.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F9b7812a1-a1a2-4e17-9bf2-88c11aac4e08_axa_2010_half_year_financial_report.pdf.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F3f4cc3bd-6823-4ccf-a918-f0c9d9063c2a_axa_us_statutory_statements_fy2010.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F5a8a399f-9a0a-4475-8fbd-5bc0ca1dffe6_axa_us_statutory_statements_3q10.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F67ba6c6d-7063-41d4-ad4e-75d86b15da43_axa_us_statutory_statements_1q10.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fa151a532-da4f-4d12-8b3b-9867df4f9724_axa_annual+financial+report_2009.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F5f89c4dd-d935-47fe-ac69-23fada9bfc96_axa_2009_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ff322c77a-e2a2-4cd7-88a0-edd8ad4cd021_axa_annual+financial+report_2008.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fb7f88f05-053a-460b-aa4d-6163d3644cfc_axa_activity_csr_report_2008_vad.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ff657a419-e066-485a-a58e-1d2870a6a035_axa_2008_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F820b669d-b3b5-4c14-986d-2223e2bcbcfb_axa_annual+financial+report_2007.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F2741b55e-9349-47ef-9704-3cbca0853b76_axa_activity_csr_report_2007.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F42159571-a3f1-4d36-b4b9-a5493fcc95e3_axa_2007_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F40f9da2a-1bcb-4e5e-9380-18f64b3ce86e_axa_annual+financial+report_2006b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fa6a14e0c-62cd-4812-a2d0-3a0aae8c862d_axa_activity_csr_report_2006b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Faf242b66-1308-4331-829f-fa91bd0db43e_axa_annual+financial+report_2005.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F6b3313d1-3b72-4f28-bc7b-f445b9b3190c_axa_activity_csr_report_2005.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F048b0d90-b28f-4fc3-bc30-b02cf8e0d6fc_axa_annual+financial+report_2004_ci.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F02acbd05-712f-4b73-93f0-dffa37e2faa2_axa_annual+financial+report_2004_ci.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fcf0b84a5-6da9-499d-985f-530559940494_axa_activity_csr_report_2004.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fafa397b5-d613-40f3-a28f-81bde0d461e2_axa_annual+financial+report_2003.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F2a31ebb9-ba04-4998-982e-9dd336abca1f_axa_annual+financial+report_2002.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F51e5f017-954b-4f81-84f9-15a086bf1e33_axa_annual+financial+report_2002_ci01.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F05fea38c-c626-4aaf-9ead-10e9c8f849c1_axa_annual+financial+report_2002_ci02.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F3e41d00d-42b3-4bfd-babc-8b9f76b73d95_axa_activity_csr_report_2002.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F4f2676f4-d36c-4d2e-b088-ef26878ff28b_axa_annual+financial+report_2001.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ffdfa0941-6fb5-4ce8-9f42-3b0152e72ce2_axa_activity_csr_report_2001.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F98922150-f1c5-4df4-9006-a8ef17a514cd_axa_annual+financial+report_2000.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F1a645a94-1c56-43be-9a5a-94495e902a23_axa_activity_csr_report_2000.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F51c109ca-2bba-45b3-a03b-78fdd16faeca_axa_annual+financial+report_1999.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F21cdedc6-c082-4ae6-abb3-4c57f0cf9dd8_axa_annual+financial+report_1998.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fd3132d9d-b656-470d-ba4f-fe8d51586e4b_axa_activity_csr_report_1998.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F746d88d3-a4f7-4126-b539-a5da353f53d7_axa_annual+financial+report_1997.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F97097956-6cd5-4fb4-a6ea-9aeb32fd9023_axa_activity_csr_report_1997.pdf"
+]
+
+
+docs_tarifs= [
+    "https://www.axa.de/site/axa-de/get/documents_E1805589786/axade/medien/privatkunden/fahrzeugversicherungen/kfz-versicherung/start-and-drive/start-and-drive-versicherungsbedingungen.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E-298610932/axade/medien/privatkunden/haftpflicht-und-recht/rechtsschutz/versicherungsbedingungen-roland-rechtsschutz.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E101690225/axade/medien/privatkunden/haftpflicht-und-recht/private%20haftpflichtversicherung/privathaftpflicht-versicherungsbedingungen-leistungspaket-S-5-mio.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E-1067805129/axade/medien/privatkunden/haftpflicht-und-recht/private%20haftpflichtversicherung/privathaftpflicht-versicherungsbedingungen-leistungspaket-S-10-mio.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1026401604/axade/medien/privatkunden/haftpflicht-und-recht/private%20haftpflichtversicherung/privathaftpflicht-versicherungsbedingungen-leistungspaket-M.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1450059874/axade/medien/privatkunden/haftpflicht-und-recht/private%20haftpflichtversicherung/privathaftpflicht-versicherungsbedingungen-leistungspaket-L.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1636759799/axade/medien/privatkunden/haus-und-wohnen/hausratversicherung/hausrat-versicherungsbedingungen-S.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1147682774/axade/medien/privatkunden/haus-und-wohnen/hausratversicherung/hausrat-versicherungsbedingungen-M-20%25.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1642308493/axade/medien/privatkunden/haus-und-wohnen/hausratversicherung/hausrat-versicherungsbedingungen-M-40%25.pdf",
+    "https://www.axa.de/site/axa-de/get/documents_E1883536226/axade/medien/privatkunden/haus-und-wohnen/hausratversicherung/hausrat-versicherungsbedingungen-L.pdf",
+]
+
+docs_list = [
+    "https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/85ec0278-bf2f-4392-94b9-c086717fa8f6_axa_urd2022_accessible_va.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/e3f52b5e-d4aa-4fc8-8bcd-f432df86e804_axa_urd_2021_en_accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com/d6aef906-e41f-40c7-ac9c-29044e98939d_AXA_URD_2020_EN_accessible_b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ffd5a8bd8-9ef1-40eb-b953-c268c0ab4bf9_axa-ri2019-en-accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F0a5e0bd9-78f2-4ef8-b32c-1d3d35ddce80_axa-ri2018-en-accessible.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F01f6966b-c26c-4935-91dc-1b296511ba8c_axa_ri2017_gb_planche.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F6f3108fd-fabc-4dc6-a984-23eb0dca7a19_axa-ri2016-en_01.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fe2936c1a-65f0-40db-b34b-bef9c27e91c0_axa_2015_half_year_financial_report.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fbbb94857-f5d4-4afd-81d0-e85666883936_axa_annual+financial+report_2014.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F25fee379-c187-40e7-bf3a-5fe1423cec0f_axa_annual+financial+report_2013.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F9224097f-d703-4efd-8050-6553ef4336f8_axa_annual+financial+report_2012b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F0b75d1fe-4b11-4462-9883-4e3bc7532bf4_axa_annual+financial+report_2011.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fc76f47f4-0917-4fb1-b1ae-78e2a4fbcef5_axa_annual+financial+report_2010c+%281%29.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fa151a532-da4f-4d12-8b3b-9867df4f9724_axa_annual+financial+report_2009.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Ff322c77a-e2a2-4cd7-88a0-edd8ad4cd021_axa_annual+financial+report_2008.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F820b669d-b3b5-4c14-986d-2223e2bcbcfb_axa_annual+financial+report_2007.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F40f9da2a-1bcb-4e5e-9380-18f64b3ce86e_axa_annual+financial+report_2006b.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Faf242b66-1308-4331-829f-fa91bd0db43e_axa_annual+financial+report_2005.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F02acbd05-712f-4b73-93f0-dffa37e2faa2_axa_annual+financial+report_2004_ci.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2Fafa397b5-d613-40f3-a28f-81bde0d461e2_axa_annual+financial+report_2003.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F2a31ebb9-ba04-4998-982e-9dd336abca1f_axa_annual+financial+report_2002.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F4f2676f4-d36c-4d2e-b088-ef26878ff28b_axa_annual+financial+report_2001.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F98922150-f1c5-4df4-9006-a8ef17a514cd_axa_annual+financial+report_2000.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F51c109ca-2bba-45b3-a03b-78fdd16faeca_axa_annual+financial+report_1999.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F21cdedc6-c082-4ae6-abb3-4c57f0cf9dd8_axa_annual+financial+report_1998.pdf"
+     ,"https://www-axa-com.cdn.axa-contento-118412.eu/www-axa-com%2F746d88d3-a4f7-4126-b539-a5da353f53d7_axa_annual+financial+report_1997.pdf"
+]
+
+
+docs = load_from_web(docs_tarifs)
+sub_docs = load_and_split(docs, chunk_size=700)
+
+# %%
+create_and_add("axa_terms", sub_docs, "hkunlp/instructor-large")
+
+docs = load_from_web(docs_list)
+sub_docs = load_and_split(docs)
+
+# %%
+create_and_add("axa_gpt", sub_docs, "hkunlp/instructor-large")

app/playground/result.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5b61c0f601cb65f2779f18fdbe5bf47f88d61f23dfbe2afdafb64c951207da8
+size 429

app/playground/st_render_doc.py

@@ -0,0 +1,9 @@
+import streamlit as st
+import utils as ut
+import cloudpickle
+
+filename="./result.pkl"
+pikd = open(filename, "rb")
+result = dict( cloudpickle.load(pikd) )
+del pikd
+ut.format_result_set(result)

app/run.py

@@ -0,0 +1,59 @@
+# This script inits the models and adds an example collection to the Vectorstore 
+# %%
+import os
+import pathlib
+
+from load_model import load_embedding
+from utils import get_chroma_client
+from load_vectors import load_from_web, create_and_add, load_and_split, metadata_generator
+
+current_path = str( pathlib.Path(__file__).parent.resolve() )
+with open(current_path+'/.openaiapikey', 'r') as reader:
+    os.environ['OPENAI_API_KEY']=reader.read()
+import load_model
+
+# %%
+#load_model.load_gpu_model("decapoda-research/llama-7b-hf") #Download local model
+#llm= load_model.load_openai_model()
+
+# %% 
+#Load example Data
+client = get_chroma_client()
+client.reset()
+ef = load_embedding("hkunlp/instructor-large")
+collection_name="papers"
+metadata= {"loaded_docs":[], "Subject":"Heikos Papers", "model_name": ef.model_name}
+selected_collection = client.create_collection(collection_name, embedding_function=ef, metadata=metadata) 
+
+docs_tarifs= [
+    "https://edoc.hu-berlin.de/bitstream/handle/18452/5294/33.pdf",
+    "https://arxiv.org/pdf/1702.03556v3.pdf",
+    "https://arxiv.org/pdf/1706.03762"
+]
+
+# %%
+# Load collection to get metadata
+loaded_collection = client.get_collection(collection_name)
+model_name = loaded_collection.metadata['model_name']
+
+# %%
+
+docs = load_from_web(docs_tarifs)
+sub_docs = load_and_split(docs, chunk_size=1000)
+create_and_add(collection_name, sub_docs, model_name, metadata)
+
+
+
+# %%
+#chain = load_model.create_chain(llm, collection=collection_name, model_name=model_name, metadata=metadata)
+#result = chain({"query": "Ist mein Kinderwagen bei einem Leitungswasserschaden mitversichert?"})
+#print(result)
+#llm= load_model.load_openai_model(temperature=0.1)
+
+#llm= load_model.load_cpu_model()
+
+#meta= metadata_generator(docs[0], llm)
+# %%
+#print(meta)
+
+# %%

app/utils.py

@@ -0,0 +1,35 @@
+import streamlit as st
+import latex2markdown
+from langchain.docstore.document import Document
+import chromadb
+from chromadb.config import Settings
+import load_model
+from load_model import load_embedding
+from load_vectors import load_from_file, load_and_split, create_and_add, load_from_web
+persist_directory = load_model.persist_directory
+
+def format_document(document: Document):
+    """TODO: Implement a nice style"""
+    return document.dict()
+
+def format_result_set(result):
+    st.write(latex2markdown.LaTeX2Markdown(result["result"]).to_markdown())
+
+    agree = st.checkbox('Show source documents')
+    source_documents = result["source_documents"]
+    if agree:
+        st.write('Source Documents:')
+        for document in source_documents:
+            st.write(format_document(document))
+
+#@st.cache_resource
+def get_chroma_client():
+    return chromadb.Client(Settings(chroma_db_impl="duckdb+parquet",
+                                    persist_directory=persist_directory
+                                ))
+#@st.cache_data
+def retrieve_collections():
+    client = get_chroma_client()
+    all_collections = client.list_collections()
+    collections = tuple( [{'name': collection.name, 'model_name': collection.metadata['model_name'], "metadata": collection.metadata} for collection in all_collections] )
+    return collections

docker-compose.yaml

@@ -0,0 +1,26 @@
+version: "3.9"
+services:
+  streamlit_app:
+    image: myretrievalgpt
+    build: .
+    tty: true
+    ports:
+      - 7860:7860
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - capabilities: [gpu]
+  dev_app:
+    image: myretrievalgpt
+    tty: true
+    volumes:
+      - ./app:/app
+      - ./root:/root
+    depends_on:
+      - streamlit_app 
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - capabilities: [gpu]

requirements.txt

@@ -0,0 +1,15 @@
+#git+https://github.com/hwchase17/langchain.git
+langchain==0.0.154
+git+https://github.com/huggingface/transformers.git
+#git+https://github.com/chroma-core/chroma.git
+chromadb
+accelerate
+bitsandbytes
+InstructorEmbedding
+cloudpickle
+streamlit
+requests==2.28.0
+latex2markdown
+openai
+unstructured[local-inference]
+llama-cpp-python