File size: 8,820 Bytes
49e32ea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41ed1b7
 
49e32ea
 
 
bc459f6
49e32ea
 
 
102df35
49e32ea
 
 
 
 
aa0ad5d
 
 
 
 
102df35
71c040a
49e32ea
 
102df35
 
 
 
 
 
 
 
 
 
49e32ea
 
f6036ad
9118536
 
49e32ea
 
 
 
aa0ad5d
 
302ada4
49e32ea
 
 
 
 
 
30689f9
 
 
49e32ea
71c040a
 
 
49e32ea
 
71c040a
49e32ea
30689f9
 
 
49e32ea
 
 
 
 
 
 
ae4a7ec
49e32ea
 
 
ae4a7ec
49e32ea
 
 
 
d2ddc62
49e32ea
 
71c040a
49e32ea
 
9118536
49e32ea
71c040a
 
49e32ea
 
9118536
49e32ea
71c040a
 
49e32ea
 
 
 
71c040a
49e32ea
d2ddc62
49e32ea
 
 
 
102df35
49e32ea
d2ddc62
49e32ea
 
 
 
 
 
 
 
aa0ad5d
 
49e32ea
 
 
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
# # Load in packages

# +
import os
from typing import TypeVar
from langchain.embeddings import HuggingFaceEmbeddings, HuggingFaceInstructEmbeddings
from langchain.vectorstores import FAISS


#PandasDataFrame: type[pd.core.frame.DataFrame]
PandasDataFrame = TypeVar('pd.core.frame.DataFrame')

# Disable cuda devices if necessary
#os.environ['CUDA_VISIBLE_DEVICES'] = '-1' 

#from chatfuncs.chatfuncs import *
import chatfuncs.ingest as ing

##  Load preset embeddings and vectorstore

embeddings_name = "thenlper/gte-base"

def load_embeddings(embeddings_name = "thenlper/gte-base"):


    if embeddings_name == "hkunlp/instructor-large":
        embeddings_func = HuggingFaceInstructEmbeddings(model_name=embeddings_name,
        embed_instruction="Represent the paragraph for retrieval: ",
        query_instruction="Represent the question for retrieving supporting documents: "
        )

    else: 
        embeddings_func = HuggingFaceEmbeddings(model_name=embeddings_name)

    global embeddings

    embeddings = embeddings_func

    return embeddings

def get_faiss_store(faiss_vstore_folder,embeddings):
    import zipfile
    with zipfile.ZipFile(faiss_vstore_folder + '/' + faiss_vstore_folder + '.zip', 'r') as zip_ref:
        zip_ref.extractall(faiss_vstore_folder)

    faiss_vstore = FAISS.load_local(folder_path=faiss_vstore_folder, embeddings=embeddings)
    os.remove(faiss_vstore_folder + "/index.faiss")
    os.remove(faiss_vstore_folder + "/index.pkl")
    
    global vectorstore

    vectorstore = faiss_vstore

    return vectorstore

import chatfuncs.chatfuncs as chatf

chatf.embeddings = load_embeddings(embeddings_name)
chatf.vectorstore = get_faiss_store(faiss_vstore_folder="faiss_embedding",embeddings=globals()["embeddings"])

def docs_to_faiss_save(docs_out:PandasDataFrame, embeddings=embeddings):

    print(f"> Total split documents: {len(docs_out)}")

    print(docs_out)

    vectorstore_func = FAISS.from_documents(documents=docs_out, embedding=embeddings)


    chatf.vectorstore = vectorstore_func

    out_message = "Document processing complete"

    return out_message, vectorstore_func

 # Gradio chat

import gradio as gr

def vote(data: gr.LikeData):
    if data.liked:
        print("You upvoted this response: " + data.value)
    else:
        print("You downvoted this response: " + data.value)

block = gr.Blocks(theme = gr.themes.Base())#css=".gradio-container {background-color: black}")

with block:
    ingest_text = gr.State()
    ingest_metadata = gr.State()
    ingest_docs = gr.State()

    embeddings_state = gr.State(globals()["embeddings"])
    vectorstore_state = gr.State(globals()["vectorstore"])  

    chat_history_state = gr.State()
    instruction_prompt_out = gr.State()

    gr.Markdown("<h1><center>Lightweight PDF / web page QA bot</center></h1>")        
    
    gr.Markdown("Chat with a document (alpha). This is a small model, that can only answer specific questions that are answered in the text. It cannot give overall impressions of, or summarise the document. By default the Lambeth Borough Plan '[Lambeth 2030 : Our Future, Our Lambeth](https://www.lambeth.gov.uk/better-fairer-lambeth/projects/lambeth-2030-our-future-our-lambeth)' is loaded. If you want to talk about another document or web page, please select from the second tab. If switching topic, please click the 'Clear chat' button.\n\nWarnings: This is a public app. Please ensure that the document you upload is not sensitive is any way as other users may see it! Also, please note that LLM chatbots may give incomplete or incorrect information, so please use with care.")

    current_source = gr.Textbox(label="Current data source that is loaded into the app", value="Lambeth_2030-Our_Future_Our_Lambeth.pdf")

    with gr.Tab("Chatbot"):

        with gr.Row():
            chat_height = 550
            chatbot = gr.Chatbot(height=chat_height, avatar_images=('user.jfif', 'bot.jpg'),bubble_full_width = False)
            sources = gr.HTML(value = "Source paragraphs where I looked for answers will appear here", height=chat_height)

        with gr.Row():
            message = gr.Textbox(
                label="What's your question?",
                lines=1,
            )     
        with gr.Row():
            submit = gr.Button(value="Send message", variant="secondary", scale = 1)
            clear = gr.Button(value="Clear chat", variant="secondary", scale=0)  

        examples_set = gr.Radio(label="Examples for the Lambeth Borough Plan",
            #value = "What were the five pillars of the previous borough plan?",
            choices=["What were the five pillars of the previous borough plan?",
                "What is the vision statement for Lambeth?",
                "What are the commitments for Lambeth?",
                "What are the 2030 outcomes for Lambeth?"])

        
        current_topic = gr.Textbox(label="Feature currently disabled - Keywords related to current conversation topic.", placeholder="Keywords related to the conversation topic will appear here")
            


    with gr.Tab("Load in a different PDF file or web page to chat"):
        with gr.Accordion("PDF file", open = False):
            in_pdf = gr.File(label="Upload pdf", file_count="multiple", file_types=['.pdf'])
            load_pdf = gr.Button(value="Load in file", variant="secondary", scale=0)
        
        with gr.Accordion("Web page", open = False):
            with gr.Row():
                in_web = gr.Textbox(label="Enter webpage url")
                in_div = gr.Textbox(label="(Advanced) Webpage div for text extraction", value="p", placeholder="p")
            load_web = gr.Button(value="Load in webpage", variant="secondary", scale=0) 
        
        ingest_embed_out = gr.Textbox(label="File/webpage preparation progress")

    gr.HTML(
        "<center>Powered by Orca Mini and Langchain</a></center>"
    )

    examples_set.change(fn=chatf.update_message, inputs=[examples_set], outputs=[message])

    # Load in a pdf
    load_pdf_click = load_pdf.click(ing.parse_file, inputs=[in_pdf], outputs=[ingest_text, current_source]).\
             then(ing.text_to_docs, inputs=[ingest_text], outputs=[ingest_docs]).\
             then(docs_to_faiss_save, inputs=[ingest_docs], outputs=[ingest_embed_out, vectorstore_state]).\
             then(chatf.hide_block, outputs = [examples_set])

    # Load in a webpage
    load_web_click = load_web.click(ing.parse_html, inputs=[in_web, in_div], outputs=[ingest_text, ingest_metadata, current_source]).\
             then(ing.html_text_to_docs, inputs=[ingest_text, ingest_metadata], outputs=[ingest_docs]).\
             then(docs_to_faiss_save, inputs=[ingest_docs], outputs=[ingest_embed_out, vectorstore_state]).\
             then(chatf.hide_block, outputs = [examples_set])

    # Load in a webpage

    # Click/enter to send message action
    response_click = submit.click(chatf.get_history_sources_final_input_prompt, inputs=[message, chat_history_state, current_topic, vectorstore_state, embeddings_state], outputs=[chat_history_state, sources, instruction_prompt_out], queue=False, api_name="retrieval").\
                then(chatf.turn_off_interactivity, inputs=[message, chatbot], outputs=[message, chatbot], queue=False).\
                then(chatf.produce_streaming_answer_chatbot_ctrans, inputs=[chatbot, instruction_prompt_out], outputs=chatbot)
    response_click.then(chatf.highlight_found_text, [chatbot, sources], [sources]).\
                then(chatf.add_inputs_answer_to_history,[message, chatbot, current_topic], [chat_history_state, current_topic]).\
                then(lambda: gr.update(interactive=True), None, [message], queue=False)

    response_enter = message.submit(chatf.get_history_sources_final_input_prompt, inputs=[message, chat_history_state, current_topic, vectorstore_state, embeddings_state], outputs=[chat_history_state, sources, instruction_prompt_out], queue=False).\
                then(chatf.turn_off_interactivity, inputs=[message, chatbot], outputs=[message, chatbot], queue=False).\
                then(chatf.produce_streaming_answer_chatbot_ctrans, [chatbot, instruction_prompt_out], chatbot)    
    response_enter.then(chatf.highlight_found_text, [chatbot, sources], [sources]).\
                then(chatf.add_inputs_answer_to_history,[message, chatbot, current_topic], [chat_history_state, current_topic]).\
                then(lambda: gr.update(interactive=True), None, [message], queue=False)
    
    # Clear box
    clear.click(chatf.clear_chat, inputs=[chat_history_state, sources, message, current_topic], outputs=[chat_history_state, sources, message, current_topic])
    clear.click(lambda: None, None, chatbot, queue=False)

    chatbot.like(vote, None, None)

block.queue(concurrency_count=1).launch(debug=True)
# -