File size: 5,120 Bytes
c4d1b84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# -*- coding: utf-8 -*-
"""api.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1XRryfVWG4d_ScN5ADvlZpKmREvTJN3mg
"""

import gradio as gr
import os

from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain.chains import ConversationalRetrievalChain
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.llms import HuggingFacePipeline
from langchain.chains import ConversationChain
from langchain.memory import ConversationBufferMemory
from langchain_community.llms import HuggingFaceEndpoint

from pathlib import Path
import chromadb
from unidecode import unidecode

from transformers import AutoTokenizer
import transformers
import torch
import tqdm
import accelerate

def load_doc(file_path):
    loader = PyPDFLoader(file_path)
    pages = loader.load()
    text_splitter = RecursiveCharacterTextSplitter(chunk_size = 600, chunk_overlap = 50)
    doc_splits = text_splitter.split_documents(pages)
    return doc_splits

!pip install fpdf

splt = load_doc('data.pdf')

def initialize_database(file_path):
    # Create list of documents (when valid)
    collection_name = Path(file_path).stem
    # Fix potential issues from naming convention
    ## Remove space
    collection_name = collection_name.replace(" ","-")
    ## Limit lenght to 50 characters
    collection_name = collection_name[:50]
    ## Enforce start and end as alphanumeric character
    if not collection_name[0].isalnum():
        collection_name[0] = 'A'
    if not collection_name[-1].isalnum():
        collection_name[-1] = 'Z'
    # print('list_file_path: ', list_file_path)
    print('Collection name: ', collection_name)
    # Load document and create splits
    doc_splits = load_doc(file_path)
    # Create or load vector database
    # global vector_db
    vector_db = create_db(doc_splits, collection_name)
    return vector_db, collection_name, "Complete!"

def create_db(splits, collection_name):
    embedding = HuggingFaceEmbeddings()
    new_client = chromadb.EphemeralClient()
    vectordb = Chroma.from_documents(
        documents=splits,
        embedding=embedding,
        client=new_client,
        collection_name=collection_name,
        # persist_directory=default_persist_directory
    )
    return vectordb

vec = initialize_database('data.pdf')

vec_cre = create_db(splt, 'data')
vec_cre

vec

def initialize_llmchain(temperature, max_tokens, top_k, vector_db):
    memory = ConversationBufferMemory(
        memory_key="chat_history",
        output_key='answer',
        return_messages=True
    )

    llm = HuggingFaceEndpoint(
            repo_id='mistralai/Mixtral-8x7B-Instruct-v0.1',
            # model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k, "load_in_8bit": True},
            temperature = temperature,
            max_new_tokens = max_tokens,
            top_k = top_k,
            load_in_8bit = True
        )
    retriever=vector_db.as_retriever()
    qa_chain = ConversationalRetrievalChain.from_llm(
        llm,
        retriever=retriever,
        chain_type="stuff",
        memory=memory,
        # combine_docs_chain_kwargs={"prompt": your_prompt})
        return_source_documents=True,
        #return_generated_question=False,
        verbose=False,
    )
    return qa_chain

qa = initialize_llmchain(0.7, 1024, 1, vec_cre)

def format_chat_history(message, chat_history):
    formatted_chat_history = []
    for user_message, bot_message in chat_history:
        formatted_chat_history.append(f"User: {user_message}")
        formatted_chat_history.append(f"Assistant: {bot_message}")
    return formatted_chat_history

def conversation(message, history):
    formatted_chat_history = format_chat_history(message, history)
    #print("formatted_chat_history",formatted_chat_history)

    # Generate response using QA chain
    response = qa({"question": message, "chat_history": formatted_chat_history})
    response_answer = response["answer"]
    if response_answer.find("Helpful Answer:") != -1:
        response_answer = response_answer.split("Helpful Answer:")[-1]
    response_sources = response["source_documents"]
    response_source1 = response_sources[0].page_content.strip()
    response_source2 = response_sources[1].page_content.strip()
    response_source3 = response_sources[2].page_content.strip()
    # Langchain sources are zero-based
    response_source1_page = response_sources[0].metadata["page"] + 1
    response_source2_page = response_sources[1].metadata["page"] + 1
    response_source3_page = response_sources[2].metadata["page"] + 1
    # print ('chat response: ', response_answer)
    # print('DB source', response_sources)

    # Append user message and response to chat history
    # return gr.update(value=""), new_history, response_sources[0], response_sources[1]
    return response_answer

conversation("what is dat gov ma", "")

gr.ChatInterface(conversation).launch()