BLAB Chatbot Client: Interpreter

This repository contains an implementation of a bot that acts a conversation manager and uses an interpreter to perform three tasks:

correction: remove errors (grammar, punctuation, capitalization, etc.) and produce a self-contained sentence by replacing external references (implicit subjects, pronouns, etc.) considering the message history;
redirection: choose the most appropriate bot to answer each question or detect that none of them can deal with the topic;
completion: generate a complete sentence containing the answerer output.

For information about the message flow, see the sequence diagram.

This bot should be used with BLAB Controller.

Installation

Install a version of Python ≥ 3.10.

Create the prompt template files for correction, redirection and completion. They should be Jinja templates that produce plain text files. Examples:

Correction and completion prompts:

Consider the following message history:
{% for message in history: -%}
    {% if message.sent_by_human -%}
        USER
    {%- else -%}
        BOT
    {%- endif -%}
    : «{{ message.text }}»
{% endfor %}

Rewrite the last message [...].

Redirection prompt:

Consider the following question-answering bots:

{% for description in bots.values(): -%}
    {{ loop.index }}: «{{ description -}}»
{% endfor %}

Which one should is best suited to answer the following question?
«{{ user_message }}»

If the topic is unrelated or the message is wrong, answer 0. Send ONLY the number, without text.

Follow these installation instructions using settings_interpreter_template.py as a template.
Follow these instructions to execute the program.

IMPORTANT: Make sure that the interpreter bots do not use the message history on their own, because the previous messages are included in the prompt. For example, if the interpreter bot is uses one of the OpenAI bots, set HISTORY_SIZE to 1.

C4AI/blab-chatbot-interpreter

BLAB Chatbot Client: Interpreter

Installation