Hugging Face's logo

Spaces:
jerpint
/
RAGTheDocs-mila-qc
Sleeping

App Files Community
2
RAGTheDocs-mila-qc
File size: 4,268 Bytes 
1c29b1a
ac493ec
 
 
 
 
 
 
1c29b1a
ac493ec
 
 
 
1c29b1a
 
 
 
 
 
 
06a349e
 
1c29b1a
 
 
 
 
 
ac493ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75f72d8
1c29b1a
ac493ec
 
 
 
 
75f72d8
ac493ec
 
 
 
 
75f72d8
 
 
ac493ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import os
from typing import Optional, Tuple

import gradio as gr
import pandas as pd
from buster.completers import Completion
from buster.utils import extract_zip

from rtd_scraper.scrape_rtd import scrape_rtd
import cfg
from cfg import setup_buster


# Check if an openai key is set as an env. variable
if os.getenv("OPENAI_API_KEY") is None:
    print(
        "Warning: No openai key detected. You can set it with 'export OPENAI_API_KEY=sk-...'."
    )


homepage_url = os.getenv("READTHEDOCS_URL") # e.g. "https://orion.readthedocs.io/"
target_version = os.getenv("READTHEDOCS_VERSION") # e.g. "en/stable"

# scrape and embed content from readthedocs website
scrape_rtd(
    homepage_url=homepage_url, save_directory="outputs/", target_version=target_version
)


# Typehint for chatbot history
ChatHistory = list[list[Optional[str], Optional[str]]]

buster = setup_buster(cfg.buster_cfg)


def add_user_question(
    user_question: str, chat_history: Optional[ChatHistory] = None
) -> ChatHistory:
    """Adds a user's question to the chat history.

    If no history is provided, the first element of the history will be the user conversation.
    """
    if chat_history is None:
        chat_history = []
    chat_history.append([user_question, None])
    return chat_history


def format_sources(matched_documents: pd.DataFrame) -> str:
    if len(matched_documents) == 0:
        return ""

    matched_documents.similarity_to_answer = (
        matched_documents.similarity_to_answer * 100
    )

    # drop duplicate pages (by title), keep highest ranking ones
    matched_documents = matched_documents.sort_values(
        "similarity_to_answer", ascending=False
    ).drop_duplicates("title", keep="first")

    documents_answer_template: str = "📝 Here are the sources I used to answer your question:\n\n{documents}\n\n{footnote}"
    document_template: str = "[🔗 {document.title}]({document.url}), relevance: {document.similarity_to_answer:2.1f} %"

    documents = "\n".join(
        [
            document_template.format(document=document)
            for _, document in matched_documents.iterrows()
        ]
    )
    footnote: str = "I'm a bot 🤖 and not always perfect."

    return documents_answer_template.format(documents=documents, footnote=footnote)


def add_sources(history, completion):
    if completion.answer_relevant:
        formatted_sources = format_sources(completion.matched_documents)
        history.append([None, formatted_sources])

    return history


def chat(chat_history: ChatHistory) -> Tuple[ChatHistory, Completion]:
    """Answer a user's question using retrieval augmented generation."""

    # We assume that the question is the user's last interaction
    user_input = chat_history[-1][0]

    # Do retrieval + augmented generation with buster
    completion = buster.process_input(user_input)

    # Stream tokens one at a time to the user
    chat_history[-1][1] = ""
    for token in completion.answer_generator:
        chat_history[-1][1] += token

        yield chat_history, completion


demo = gr.Blocks()
with demo:
    with gr.Row():
        gr.Markdown("<h3><center>RAGTheDocs</center></h3>")

    chatbot = gr.Chatbot()

    with gr.Row():
        question = gr.Textbox(
            label="What's your question?",
            placeholder="Type your question here...",
            lines=1,
        )
        submit = gr.Button(value="Send", variant="secondary")

    examples = gr.Examples(
        examples=[
            "How can I install the library?",
            "What dependencies are required?",
            "Give a brief overview of the library."
        ],
        inputs=question,
    )

    gr.Markdown(
        "This app uses [Buster 🤖](github.com/jerpint/buster) and ChatGPT to search the docs for relevant info and answer questions."
    )

    response = gr.State()

    # fmt: off
    gr.on(
        triggers=[submit.click, question.submit],
        fn=add_user_question,
        inputs=[question],
        outputs=[chatbot]
    ).then(
        chat,
        inputs=[chatbot],
        outputs=[chatbot, response]
    ).then(
        add_sources,
        inputs=[chatbot, response],
        outputs=[chatbot]
    )


demo.queue(concurrency_count=16)
demo.launch(share=False)