File size: 7,583 Bytes
a5686cb
 
a4595fc
a5686cb
71ab0a8
19a9d09
 
 
 
 
fa7f0c5
669d503
99e2b1f
 
 
fdf1622
 
 
99e2b1f
 
a4595fc
99e2b1f
 
fdf1622
 
 
669d503
 
 
a4595fc
 
0b4f4a2
82532b2
0b4f4a2
 
 
 
 
 
 
 
 
 
 
 
9f6c9bd
 
 
 
 
 
 
 
99e2b1f
f3d1657
a4595fc
 
 
 
 
 
 
 
 
bf93486
68fbb90
 
 
 
 
 
121f27f
 
 
 
 
 
 
 
 
 
 
fdf1622
a422880
 
121f27f
a422880
 
a5686cb
 
dc1d7e6
 
 
 
fdf1622
 
 
 
bf93486
f3d1657
 
 
 
adfbbbe
19a9d09
 
af9539a
dc1d7e6
af9539a
c1646ce
97ba4cb
121f27f
82532b2
19a9d09
af9539a
 
 
a422880
af9539a
 
 
 
 
 
 
 
 
 
 
 
a4595fc
af9539a
 
fdf1622
af9539a
 
a4595fc
af9539a
 
 
 
 
 
 
 
fdf1622
 
19a9d09
 
dc1d7e6
 
 
19a9d09
 
af9539a
 
 
 
 
 
a4595fc
 
a5686cb
af9539a
19a9d09
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4dd3ec8
d730458
 
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
191
192
193
194
import gradio as gr
from haystack.document_stores import FAISSDocumentStore
from haystack.nodes import EmbeddingRetriever
import openai
import os
from utils import (
    make_pairs,
    set_openai_api_key,
    get_random_string,
)

system_template = {"role": "system", "content": os.environ["content"]}

retrieve_all = EmbeddingRetriever(
    document_store=FAISSDocumentStore.load(
        index_path="./documents/climate_gpt.faiss",
        config_path="./documents/climate_gpt.json",
    ),
    embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
    model_format="sentence_transformers",
)
retrieve_giec = EmbeddingRetriever(
    document_store=FAISSDocumentStore.load(
        index_path="./documents/climate_gpt_only_giec.faiss",
        config_path="./documents/climate_gpt_only_giec.json",
    ),
    embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
    model_format="sentence_transformers",
)


def chat(
    query: str, history: list = [system_template], report_type: str = "All available", threshold: float = 0.559
) -> tuple:
    """retrieve relevant documents in the document store then query gpt-turbo

    Args:
        query (str): user message.
        history (list, optional): history of the conversation. Defaults to [system_template].
        report_type (str, optional): should be "All available" or "IPCC only". Defaults to "All available".
        threshold (float, optional): similarity threshold, don't increase more than 0.568. Defaults to 0.56.

    Yields:
        tuple: chat gradio format, chat openai format, sources used.
    """

    if report_type == "All available":
        retriever = retrieve_all
    elif report_type == "IPCC only":
        retriever = retrieve_giec
    else:
        raise Exception("report_type arg should be in (All available, IPCC only)")

    docs = retriever.retrieve(query=query, top_k=10)

    messages = history + [{"role": "user", "content": query}]
    sources = "\n\n".join(
        f"doc {i}: {d.meta['file_name']} page {d.meta['page_number']}\n{d.content}"
        for i, d in enumerate(docs, 1)
        if d.score > threshold
    )

    if sources:
        messages.append({"role": "system", "content": f"{os.environ['sources']}\n\n{sources}"})

    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=messages,
        temperature=0.2,
        stream=True,
    )

    if sources:
        messages.pop()
        complete_response = ""
    else:
        sources = "No environmental report was used to provide this answer."
        complete_response = (
            "No relevant documents found, for a sourced answer you may want to try a more specific question.\n\n"
        )

    messages.append({"role": "assistant", "content": complete_response})
    for chunk in response:
        if chunk_message := chunk["choices"][0]["delta"].get("content", None):
            complete_response += chunk_message
            messages[-1]["content"] = complete_response
            gradio_format = make_pairs([a["content"] for a in messages[1:]])
            yield gradio_format, messages, sources


def test(feed: str):
    print(feed)


def reset_textbox():
    return gr.update(value="")


# 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:

    openai.api_key = os.environ["api_key"]
    user_id = gr.State([get_random_string(10)])

    with gr.Tab("App"):
        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 other environmental reports.
            \n **How does it work:** when a user sends a message, the system retrieves the most relevant paragraphs from scientific reports that are semantically related to the user's question. These paragraphs are then used to generate a comprehensive and well-sourced answer using a language model.
            \n **Usage guideline:** more sources will be retrieved using precise questions.
            \n ⚠️ Always refer to the source to ensure the validity of the information communicated.
            """
        )
        with gr.Row():
            with gr.Column(scale=2):
                chatbot = gr.Chatbot(elem_id="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=chat,
            inputs=[
                ask,
                state,
                gr.inputs.Dropdown(
                    ["IPCC only", "All available"],
                    default="All available",
                    label="Select reports",
                ),
            ],
            outputs=[chatbot, state, sources_textbox],
        )
        ask.submit(reset_textbox, [], [ask])

        with gr.Accordion("Feedbacks", open=False):
            gr.Markdown("Please complete some feedbacks 🙏")
            feedback = gr.Textbox()
            feedback_save = gr.Button(value="submit feedback")
            feedback_save.click(test, inputs=[feedback])

        with gr.Accordion("Add your personal openai api key - Option", 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])

    with gr.Tab("Information"):
        gr.Markdown(
            """
        ## 📖 Reports used : \n
        - First Assessment Report on the Physical Science of Climate Change
        - Second assessment Report on Climate Change Adaptation
        - Third Assessment Report on Climate Change Mitigation
        - Food Outlook Biannual Report on Global Food Markets
        - IEA's report on the Role of Critical Minerals in Clean Energy Transitions
        - Limits to Growth
        - Outside The Safe operating system of the Planetary Boundary for Novel Entities
        - Planetary Boundaries Guiding
        - State of the Oceans report
        - Word Energy Outlook 2021
        - Word Energy Outlook 2022
        - The environmental impacts of plastics and micro plastics use, waste and polution ET=U and national measures
        - IPBES Global report - MArch 2022

        \n
        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.

        """
        )
    with gr.Tab("Examples"):
        gr.Markdown("See here some examples on how to use the Chatbot")

    demo.queue(concurrency_count=16)

demo.launch()