File size: 6,208 Bytes
2efdfac
749b606
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
faf225d
749b606
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr
import paperqa
import pickle
import pandas as pd
from pathlib import Path
import requests
import zipfile
import io
import tempfile
import os


css_style = """
.gradio-container {
    font-family: "IBM Plex Mono";
}
"""


def request_pathname(files, data, openai_api_key):
    if files is None:
        return [[]]
    for file in files:
        # make sure we're not duplicating things in the dataset
        if file.name in [x[0] for x in data]:
            continue
        data.append([file.name, None, None])
    return [[len(data), 0]], data, data, validate_dataset(pd.DataFrame(data), openai_api_key)


def validate_dataset(dataset, openapi):
    docs_ready = dataset.iloc[-1, 0] != ""
    if docs_ready and type(openapi) is str and len(openapi) > 0:
        return "✨Ready✨"
    elif docs_ready:
        return "⚠️Waiting for key⚠️"
    elif type(openapi) is str and len(openapi) > 0:
        return "⚠️Waiting for documents⚠️"
    else:
        return "⚠️Waiting for documents and key⚠️"


def make_stats(docs):
    return [[len(docs.doc_previews), sum([x[0] for x in docs.doc_previews])]]


# , progress=gr.Progress()):
def do_ask(question, button, openapi, dataset, length, do_marg, k, max_sources, docs):
    passages = ""
    docs_ready = dataset.iloc[-1, 0] != ""
    if button == "✨Ready✨" and type(openapi) is str and len(openapi) > 0 and docs_ready:
        os.environ['OPENAI_API_KEY'] = openapi.strip()
        if docs is None:
            docs = paperqa.Docs()
        # dataset is pandas dataframe
        for _, row in dataset.iterrows():
            try:
                docs.add(row['filepath'], row['citation string'],
                         key=row['key'], disable_check=True)
                yield "", "", "", docs, make_stats(docs)
            except Exception as e:
                pass
    else:
        yield "", "", "", docs, [[0, 0]]
    #progress(0, "Building Index...")
    docs._build_faiss_index()
    #progress(0.25, "Querying...")
    for i, result in enumerate(docs.query_gen(question,
                                              length_prompt=f'use {length:d} words',
                                              marginal_relevance=do_marg,
                                              k=k, max_sources=max_sources)):
        #progress(0.25 + 0.1 * i, "Generating Context" + str(i))
        yield result.formatted_answer, result.context, passages, docs,  make_stats(docs)
    #progress(1.0, "Done!")
    # format the passages
    for i, (key, passage) in enumerate(result.passages.items()):
        passages += f'Disabled for now'
    yield result.formatted_answer, result.context, passages, docs,  make_stats(docs)




with gr.Blocks(css=css_style) as demo:

    docs = gr.State(None)
    data = gr.State([])
    openai_api_key = gr.State('')

    gr.Markdown(f"""
    # Document Question and Answer (v{paperqa.__version__})
    *By Andrew White ([@andrewwhite01](https://twitter.com/andrewwhite01))*
    This tool will enable asking questions of your uploaded text, PDF documents,
    or scrape github repos.
    It uses OpenAI's GPT models and thus you must enter your API key below. This
    tool is under active development and currently uses many tokens - up to 10,000
    for a single query. That is $0.10-0.20 per query, so please be careful!
    * [PaperQA](https://github.com/whitead/paper-qa) is the code used to build this tool.
    * [langchain](https://github.com/hwchase17/langchain) is the main library this tool utilizes.
    1. Enter API Key ([What is that?](https://platform.openai.com/account/api-keys))
    2. Upload your documents
    3. Ask a questions
    """)
    openai_api_key = gr.Textbox(
        label="OpenAI API Key", placeholder="sk-...", type="password")
    with gr.Tab('File Upload'):
        uploaded_files = gr.File(
            label="Your Documents Upload (PDF or txt)", file_count="multiple", )
    
    with gr.Accordion("See Docs:", open=False):
        dataset = gr.Dataframe(
            headers=["filepath", "citation string", "key"],
            datatype=["str", "str", "str"],
            col_count=(3, "fixed"),
            interactive=False,
            label="Documents and Citations",
            overflow_row_behaviour='paginate',
            max_rows=5
        )
    buildb = gr.Textbox("⚠️Waiting for documents and key...",
                        label="Status", interactive=False, show_label=True,
                        max_lines=1)
    stats = gr.Dataframe(headers=['Docs', 'Chunks'],
                         datatype=['number', 'number'],
                         col_count=(2, "fixed"),
                         interactive=False,
                         label="Doc Stats")
    openai_api_key.change(validate_dataset, inputs=[
                          dataset, openai_api_key], outputs=[buildb])
    dataset.change(validate_dataset, inputs=[
                   dataset, openai_api_key], outputs=[buildb])
    uploaded_files.change(request_pathname, inputs=[
                          uploaded_files, data, openai_api_key], outputs=[stats, data, dataset, buildb])
    
    query = gr.Textbox(
        placeholder="Enter your question here...", label="Question")
    with gr.Row():
        length = gr.Slider(25, 200, value=100, step=5,
                           label='Words in answer')
        marg = gr.Checkbox(True, label='Max marginal relevance')
        k = gr.Slider(1, 20, value=10, step=1,
                      label='Chunks to examine')
        sources = gr.Slider(1, 10, value=5, step=1,
                            label='Contexts to include')

    ask = gr.Button("Ask Question")
    answer = gr.Markdown(label="Answer")
    with gr.Accordion("Context", open=True):
        context = gr.Markdown(label="Context")

    with gr.Accordion("Raw Text", open=False):
        passages = gr.Markdown(label="Passages")
    ask.click(fn=do_ask, inputs=[query, buildb,
                                 openai_api_key, dataset,
                                 length, marg, k, sources,
                                 docs], outputs=[answer, context, passages, docs, stats])

demo.queue(concurrency_count=20)
demo.launch(show_error=True)