File size: 5,668 Bytes
a5686cb
 
 
 
 
 
71ab0a8
bf93486
 
 
 
fa7f0c5
a5686cb
6a89c2d
fa7f0c5
a5686cb
 
f3d1657
bf93486
a5686cb
bf93486
 
 
 
a5686cb
bf93486
 
 
f3d1657
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bf93486
a5686cb
 
 
 
 
 
 
 
 
5cba42a
a5686cb
6a89c2d
a5686cb
 
 
 
 
 
 
 
 
 
 
 
 
 
6a89c2d
a5686cb
6a89c2d
a5686cb
 
bf93486
 
 
a5686cb
 
 
 
 
 
bf93486
f3d1657
 
 
 
 
 
 
 
a5686cb
f3d1657
 
 
 
 
a327c68
adfbbbe
bf93486
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a5686cb
 
0fe1b3a
a5686cb
 
 
 
bf93486
 
 
a5686cb
bf93486
a5686cb
 
 
 
 
 
bf93486
a5686cb
bf93486
f3d1657
bf93486
a5686cb
12f6ef4
 
 
 
 
 
 
2983354
 
a5686cb
f9aee46
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
import gradio as gr
from transformers import pipeline
from haystack.document_stores import FAISSDocumentStore
from haystack.nodes import EmbeddingRetriever
import numpy as np
import openai
import os
from datasets import load_dataset
from datasets import Dataset
import time
from utils import is_climate_change_related, make_pairs, set_openai_api_key


classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
system_template = {"role": os.environ["role"], "content": os.environ["content"]}


def gen_conv(query: str, report_type, history=[system_template], ipcc=True):
    """return (answer:str, history:list[dict], sources:str)

    Args:
        query (str): _description_
        history (list, optional): _description_. Defaults to [system_template].
        ipcc (bool, optional): _description_. Defaults to True.

    Returns:
        _type_: _description_
    """
    if report_type == "giec":
        document_store = FAISSDocumentStore.load(
            index_path="./documents/climate_gpt_only_giec.faiss",
            config_path="./documents/climate_gpt_only_giec.json",
        )
    else:
        document_store = FAISSDocumentStore.load(
            index_path="./documents/climate_gpt.faiss",
            config_path="./documents/climate_gpt.json",
        )

    dense = EmbeddingRetriever(
        document_store=document_store,
        embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
        model_format="sentence_transformers",
    )

    retrieve = ipcc and is_climate_change_related(query, classifier)

    sources = ""
    messages = history + [
        {"role": "user", "content": query},
    ]

    if retrieve:
        docs = dense.retrieve(query=query, top_k=5)
        sources = "\n\n".join(
            [os.environ["sources"]]
            + [
                f"{d.meta['file_name']} Page {d.meta['page_number']}\n{d.content}"
                for d in docs
            ]
        )
        messages.append({"role": "system", "content": sources})

    answer = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=messages,
        temperature=0.2,
        #         max_tokens=200,
    )["choices"][0]["message"]["content"]

    if retrieve:
        messages.pop()
        # answer = "(top 5 documents retrieved) " + answer
        sources = "\n\n".join(
            f"{d.meta['file_name']} Page {d.meta['page_number']}:\n{d.content}"
            for d in docs
        )
    else:
        sources = "No environmental report was used to provide this answer."

    messages.append({"role": "assistant", "content": answer})
    gradio_format = make_pairs([a["content"] for a in messages[1:]])

    return gradio_format, messages, sources


# Gradio
css_code = ".gradio-container {background-image: url('file=background.png');background-position: top right}"

with gr.Blocks(title="🌍 ClimateGPT Ekimetrics", css=css_code) as demo:

    document_store = FAISSDocumentStore.load(
        index_path="./documents/climate_gpt.faiss",
        config_path="./documents/climate_gpt.json",
    )

    dense = EmbeddingRetriever(
        document_store=document_store,
        embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
        model_format="sentence_transformers",
    )

    openai.api_key = os.environ["api_key"]
    gr.Markdown("### Welcome to Climate GPT 🌍 ! ")
    gr.Markdown(
        """
        Climate GPT is an interactive exploration tool designed to help you easily find relevant information based on  of Environmental reports such as IPCCs and ??.

        IPCC is a United Nations body that assesses the science related to climate change, including its impacts and possible response options. The IPCC is considered the leading scientific authority on all things related to global climate change.
    """
    )
    gr.Markdown(
        "**How does it work:** This Chatbot is a combination of two technologies. FAISS search applied to a vast amount of scientific climate reports and TurboGPT to generate human-like text from the part of the document extracted from the database."
    )
    gr.Markdown(
        "⚠️ Warning: Always refer to the source (on the right side) to ensure the validity of the information communicated"
    )
    # gr.Markdown("""### Ask me anything, I'm a climate expert""")
    with gr.Row():
        with gr.Column(scale=2):
            chatbot = gr.Chatbot()
            state = gr.State([system_template])

            with gr.Row():
                ask = gr.Textbox(
                    show_label=False,
                    placeholder="Enter text and press enter",
                    sample_inputs=["which country polutes the most ?"],
                ).style(container=False)
                print(f"Type from ask textbox {ask.type}")

        with gr.Column(scale=1, variant="panel"):
            gr.Markdown("### Sources")
            sources_textbox = gr.Textbox(
                interactive=False, show_label=False, max_lines=50
            )

    ask.submit(
        fn=gen_conv,
        inputs=[ask, gr.inputs.Dropdown(["giec only", "all"], default="all"), state],
        outputs=[chatbot, state, sources_textbox],
    )
    with gr.Accordion("Add your personal openai api key", open=False):
        openai_api_key_textbox = gr.Textbox(
            placeholder="Paste your OpenAI API key (sk-...) and hit Enter",
            show_label=False,
            lines=1,
            type="password",
        )
    openai_api_key_textbox.change(set_openai_api_key, inputs=[openai_api_key_textbox])
    openai_api_key_textbox.submit(set_openai_api_key, inputs=[openai_api_key_textbox])

demo.launch()