app.py

import gradio as gr
import time
import yaml
from langchain.prompts.chat import ChatPromptTemplate
from huggingface_hub import hf_hub_download
from spinoza_project.source.backend.llm_utils import (
    get_llm,
    get_llm_api,
    get_vectorstore,
    get_vectorstore_api,
)
from spinoza_project.source.backend.document_store import pickle_to_document_store
from spinoza_project.source.backend.get_prompts import get_qa_prompts
from spinoza_project.source.frontend.utils import (
    make_html_source,
    make_html_presse_source,
    make_html_afp_source,
    make_html_politique_source,
    parse_output_llm_with_sources,
    init_env,
)
from spinoza_project.source.backend.prompt_utils import (
    to_chat_instruction,
    SpecialTokens,
)

from assets.utils_javascript import (
    accordion_trigger,
    accordion_trigger_end,
    accordion_trigger_spinoza,
    accordion_trigger_spinoza_end,
    update_footer,
)

init_env()

with open("./spinoza_project/config.yaml") as f:
    config = yaml.full_load(f)

prompts = {}
for source in config["prompt_naming"]:
    with open(f"./spinoza_project/prompt_{source}.yaml") as f:
        prompts[source] = yaml.full_load(f)

## Building LLM
print("Building LLM")
model = "gpt35turbo"
llm = get_llm_api()

## Loading_tools
print("Loading Databases")
bdd_presse = get_vectorstore_api("presse")
bdd_afp = get_vectorstore_api("afp")
qdrants = {
    tab: pickle_to_document_store(
        hf_hub_download(
            repo_id="TestSpinoza/spinoza-database",
            filename=f"database_{tab}.pickle",
            repo_type="dataset",
            force_download=True,
        )
    )
    for tab in config["prompt_naming"]
    if tab != "Presse" and tab != "AFP"
}

## Load Prompts
print("Loading Prompts")
chat_qa_prompts, chat_reformulation_prompts, chat_summarize_memory_prompts = {}, {}, {}
for source, prompt in prompts.items():
    chat_qa_prompt, chat_reformulation_prompt = get_qa_prompts(config, prompt)
    chat_qa_prompts[source] = chat_qa_prompt
    chat_reformulation_prompts[source] = chat_reformulation_prompt


with open("./assets/style.css", "r") as f:
    css = f.read()


special_tokens = SpecialTokens(config)

synthesis_template = """You are a factual journalist that summarize the secialized awnsers from thechnical sources.

Based on the folowing question:
{question}

And the following expert answer:
{answers}

- When using legal answers, keep tracking of the name of the articles.
- When using ADEME answers, name the sources that are mainly used.
- List the different elements mentionned, and highlight the agreement points between the sources, as well as the contradictions or differences.
- Contradictions don't lie in whether or not a subject is dealt with, but more in the opinion given or the way the subject is dealt with.
- Generate the answer as markdown, with an aerated layout, and headlines in bold
- When you use information from a passage, mention where it came from by using [Doc i] at the end of the sentence. i stands for the number of the document.",
- Do not use the sentence 'Doc i says ...' to say where information came from.",
- If the same thing is said in more than one document, you can mention all of them like this: [Doc i, Doc j, Doc k]",
- Start by highlighting contradictions, then do a general summary and finally get into the details that might be interesting for article writing. Where relevant, quote them.
- Awnser in French / Répond en Français
"""

synthesis_prompt = to_chat_instruction(synthesis_template, special_tokens)
synthesis_prompt_template = ChatPromptTemplate.from_messages([synthesis_prompt])


def zip_longest_fill(*args, fillvalue=None):
    # zip_longest('ABCD', 'xy', fillvalue='-') --> Ax By C- D-
    iterators = [iter(it) for it in args]
    num_active = len(iterators)
    if not num_active:
        return

    cond = True
    fillvalues = [None] * len(iterators)
    while cond:
        values = []
        for i, it in enumerate(iterators):
            try:
                value = next(it)
            except StopIteration:
                value = fillvalues[i]
            values.append(value)

        new_cond = False
        for i, elt in enumerate(values):
            if elt != fillvalues[i]:
                new_cond = True
        cond = new_cond

        fillvalues = values.copy()
        yield tuple(values)


def format_question(question):
    return f"{question}"  # ###


def parse_question(question):
    x = question.replace("<p>", "").replace("</p>\n", "")
    if "### " in x:
        return x.split("### ")[1]
    return x


def reformulate(question, tab, config=config):
    if tab in list(config["tabs"].keys()):
        return llm.stream(
            chat_reformulation_prompts[config["source_mapping"][tab]],
            {"question": parse_question(question)},
        )
    else:
        return iter([None] * 5)


def reformulate_single_question(question, tab, config=config):
    for elt in reformulate(question, tab, config=config):
        time.sleep(0.02)
        yield elt


def reformulate_questions(question, config=config):
    for elt in zip_longest_fill(
        *[reformulate(question, tab, config=config) for tab in config["tabs"]]
    ):
        time.sleep(0.02)
        yield elt


def add_question(question):
    return question


def answer(question, source, tab, config=config):
    if tab in list(config["tabs"].keys()):
        if len(source) < 10:
            return iter(["Aucune source trouvée, veuillez reformuler votre question"])
        else:

            return llm.stream(
                chat_qa_prompts[config["source_mapping"][tab]],
                {
                    "question": parse_question(question),
                    "sources": source.replace("<p>", "").replace("</p>\n", ""),
                },
            )
    else:
        return iter([None] * 5)


def answer_single_question(source, question, tab, config=config):
    for elt in answer(question, source, tab, config=config):
        time.sleep(0.02)
        yield elt


def answer_questions(*questions_sources, config=config):
    questions = [elt for elt in questions_sources[: len(questions_sources) // 2]]
    sources = [elt for elt in questions_sources[len(questions_sources) // 2 :]]

    for elt in zip_longest_fill(
        *[
            answer(question, source, tab, config=config)
            for question, source, tab in zip(questions, sources, config["tabs"])
        ]
    ):
        time.sleep(0.02)
        yield [
            [(question, parse_output_llm_with_sources(ans))]
            for question, ans in zip(questions, elt)
        ]


def get_sources(
    questions, qdrants=qdrants, bdd_presse=bdd_presse, bdd_afp=bdd_afp, config=config
):
    k = config["num_document_retrieved"]
    min_similarity = config["min_similarity"]
    text, formated = [], []
    for i, (question, tab) in enumerate(zip(questions, list(config["tabs"].keys()))):
        if tab == "Presse":
            sources = bdd_presse.similarity_search_with_relevance_scores(
                question.replace("<p>", "").replace("</p>\n", ""), k=k
            )
            sources = [
                (doc, score) for doc, score in sources if score >= min_similarity
            ]
            formated.extend(
                [
                    make_html_presse_source(source[0], j, source[1])
                    for j, source in zip(range(k * i + 1, k * (i + 1) + 1), sources)
                ]
            )

        elif tab == "AFP":
            sources = bdd_afp.similarity_search_with_relevance_scores(
                question.replace("<p>", "").replace("</p>\n", ""), k=k
            )
            sources = [
                (doc, score) for doc, score in sources if score >= min_similarity
            ]
            formated.extend(
                [
                    make_html_afp_source(source[0], j, source[1])
                    for j, source in zip(range(k * i + 1, k * (i + 1) + 1), sources)
                ]
            )

        elif tab == "Documents Stratégiques":
            sources = qdrants[
                config["source_mapping"][tab]
            ].similarity_search_with_relevance_scores(
                config["query_preprompt"]
                + question.replace("<p>", "").replace("</p>\n", ""),
                k=k,
            )
            sources = [
                (doc, score) for doc, score in sources if score >= min_similarity
            ]
            formated.extend(
                [
                    make_html_politique_source(source[0], j, source[1], config)
                    for j, source in zip(range(k * i + 1, k * (i + 1) + 1), sources)
                ]
            )

        else:
            sources = qdrants[
                config["source_mapping"][tab]
            ].similarity_search_with_relevance_scores(
                config["query_preprompt"]
                + question.replace("<p>", "").replace("</p>\n", ""),
                k=k,
            )
            sources = [
                (doc, score) for doc, score in sources if score >= min_similarity
            ]
            formated.extend(
                [
                    make_html_source(source[0], j, source[1], config)
                    for j, source in zip(range(k * i + 1, k * (i + 1) + 1), sources)
                ]
            )

        text.extend(
            [
                "\n\n".join(
                    [
                        f"Doc {str(j)} with source type {source[0].metadata.get('file_source_type')}:\n"
                        + source[0].page_content
                        for j, source in zip(range(k * i + 1, k * (i + 1) + 1), sources)
                    ]
                )
            ]
        )

    formated = "".join(formated)

    return formated, text


def retrieve_sources(
    *questions, qdrants=qdrants, bdd_presse=bdd_presse, bdd_afp=bdd_afp, config=config
):
    formated_sources, text_sources = get_sources(
        questions, qdrants, bdd_presse, bdd_afp, config
    )

    return (formated_sources, *text_sources)


def get_synthesis(question, *answers, config=config):
    answer = []
    for i, tab in enumerate(config["tabs"]):
        if len(str(answers[i])) >= 100:
            answer.append(
                f"{tab}\n{answers[i]}".replace("<p>", "").replace("</p>\n", "")
            )

    if len(answer) == 0:
        return "Aucune source n'a pu être identifiée pour répondre, veuillez modifier votre question"
    else:
        for elt in llm.stream(
            synthesis_prompt_template,
            {
                "question": question.replace("<p>", "").replace("</p>\n", ""),
                "answers": "\n\n".join(answer),
            },
        ):
            time.sleep(0.01)
            yield [(question, parse_output_llm_with_sources(elt))]


theme = gr.themes.Base(
    primary_hue="blue",
    secondary_hue="red",
    font=[gr.themes.GoogleFont("Poppins"), "ui-sans-serif", "system-ui", "sans-serif"],
)

with open("./assets/style.css", "r") as f:
    css = f.read()

with open("./assets/source_information.md", "r") as f:
    source_information = f.read()


def start_agents():
    gr.Info(message="The agents and Spinoza are loading...", duration=3)

    return [
        (None, "I am waiting until all the agents are done to generate an answer...")
    ]


def end_agents():
    gr.Info(
        message="The agents and Spinoza have finished answering your question",
        duration=3,
    )


def next_call():
    return


init_prompt = """
Hello, I am Spinoza, a conversational assistant designed to help you in your journalistic journey. I will answer your questions based **on the provided sources**.

⚠️ Limitations
*Please note that this chatbot is in an early stage, it is not perfect and may sometimes give irrelevant answers. If you are not satisfied with the answer, please ask a more specific question or report your feedback to help us improve the system.*

What do you want to learn ?
"""

with gr.Blocks(
    title=f"🔍 Spinoza",
    css=css,
    js=update_footer(),
    theme=theme,
) as demo:
    chatbots = {}
    question = gr.State("")
    docs_textbox = gr.State([""])
    agent_questions = {elt: gr.State("") for elt in config["tabs"]}
    component_sources = {elt: gr.State("") for elt in config["tabs"]}
    text_sources = {elt: gr.State("") for elt in config["tabs"]}
    tab_states = {elt: gr.State(elt) for elt in config["tabs"]}

    with gr.Tab("Q&A", elem_id="main-component"):
        with gr.Row(elem_id="chatbot-row"):
            with gr.Column(scale=2, elem_id="center-panel"):
                with gr.Group(elem_id="chatbot-group"):
                    with gr.Accordion(
                        "Science agent",
                        open=False,
                        elem_id="accordion-science",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[0]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-science",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "Law agent",
                        open=False,
                        elem_id="accordion-legal",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[1]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-legal",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "Politics agent",
                        open=False,
                        elem_id="accordion-politique",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[2]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-politique",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "ADEME agent",
                        open=False,
                        elem_id="accordion-ademe",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[3]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-ademe",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "Press agent",
                        open=False,
                        elem_id="accordion-presse",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[4]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-presse",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "AFP agent",
                        open=False,
                        elem_id="accordion-afp",
                        elem_classes="accordion",
                    ):
                        chatbots[list(config["tabs"].keys())[5]] = gr.Chatbot(
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-afp",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                None,
                            ),
                        )

                    with gr.Accordion(
                        "Spinoza",
                        open=True,
                        elem_id="accordion-spinoza",
                        elem_classes="accordion",
                    ):
                        chatbots["spinoza"] = gr.Chatbot(
                            value=[(None, init_prompt)],
                            show_copy_button=True,
                            show_share_button=False,
                            show_label=False,
                            elem_id="chatbot-spinoza",
                            layout="panel",
                            avatar_images=(
                                "./assets/logos/help.png",
                                "./assets/logos/spinoza.png",
                            ),
                        )

                with gr.Row(elem_id="input-message"):
                    ask = gr.Textbox(
                        placeholder="Ask me anything here!",
                        show_label=False,
                        scale=7,
                        lines=1,
                        interactive=True,
                        elem_id="input-textbox",
                    )

            with gr.Column(scale=1, variant="panel", elem_id="right-panel"):
                with gr.TabItem("Sources", elem_id="tab-sources", id=0):
                    sources_textbox = gr.HTML(
                        show_label=False, elem_id="sources-textbox"
                    )

    with gr.Tab("Source information", elem_id="source-component"):
        with gr.Row():
            with gr.Column(scale=1):
                gr.Markdown(source_information)

    with gr.Tab("Contact", elem_id="contact-component"):
        with gr.Row():
            with gr.Column(scale=1):
                gr.Markdown("For any issue contact **[email protected]**.")

    ask.submit(
        start_agents, inputs=[], outputs=[chatbots["spinoza"]], js=accordion_trigger()
    ).then(
        fn=reformulate_questions,
        inputs=[ask],
        outputs=[agent_questions[tab] for tab in config["tabs"]],
    ).then(
        fn=retrieve_sources,
        inputs=[agent_questions[tab] for tab in config["tabs"]],
        outputs=[sources_textbox] + [text_sources[tab] for tab in config["tabs"]],
    ).then(
        fn=answer_questions,
        inputs=[agent_questions[tab] for tab in config["tabs"]]
        + [text_sources[tab] for tab in config["tabs"]],
        outputs=[chatbots[tab] for tab in config["tabs"]],
    ).then(
        fn=next_call, inputs=[], outputs=[], js=accordion_trigger_end()
    ).then(
        fn=next_call, inputs=[], outputs=[], js=accordion_trigger_spinoza()
    ).then(
        fn=get_synthesis,
        inputs=[agent_questions[list(config["tabs"].keys())[1]]]
        + [chatbots[tab] for tab in config["tabs"]],
        outputs=[chatbots["spinoza"]],
    ).then(
        fn=next_call, inputs=[], outputs=[], js=accordion_trigger_spinoza_end()
    ).then(
        fn=end_agents, inputs=[], outputs=[]
    )


if __name__ == "__main__":
    demo.queue().launch(debug=True)