Skip to main content

AI Smart Human Assistant Library

Project description

AISHA Lib: A High-Level Abstraction for Building AI Assistants

In the evolving landscape of artificial intelligence, the development of smart assistants has become increasingly prevalent. To streamline this process, the AISHA (AI Smart Human Assistant) Lib offers a high-level abstraction designed for creating AI assistants. This versatile library supports various large language models (LLMs) and different LLM backends, providing developers with a powerful and flexible toolset.

Environment

To create a Python virtual environment, use the command:

conda env create -f environment.yml

Installation

pip install aishalib

Supported Models

The following LLM models are supported:

  • Phi-3-medium-128k-instruct
  • c4ai-command-r-v01

LLM backends

The following LLM backends are supported:

  • Llama.cpp Server API

Telegram bot example

from aishalib.aishalib import Aisha
from telegram import Update
from telegram.ext import Application, MessageHandler, ContextTypes, filters

TG_TOKEN = "YOUR_TG_TOKEN"

SYSTEM_PROMPT = """
Ты умный бот помощник для общения в телеграме.
Ты общаешься в групповом чате с другими пользователями.
Ты отвечаешь на русском языке.
"""

SYSTEM_INJECTION = """
Последнее сообщение написал пользователь с идентификатором {user_id}.
Используй эти идентификаторы для того, чтобы различать пользователей.
Запрещено обращаться к пользователю по его идентификатору! Можно только по имени.
Если пользователь не представился спроси как его зовут.
Если исходя из контекста и смысла беседы это сообщение адресовано тебе или это общее сообщение для всех в чате то ты обязан на него ответить.
Если это сообщение адресовано другому пользователю, то напиши специальную команду "ignoring_message" в ответе.
"""

def get_aisha(chat_id, tg_context):
    if chat_id not in tg_context.user_data:
        aisha = Aisha("http://127.0.0.1:8000/completion",
                      "CohereForAI/c4ai-command-r-v01",
                      prompt=SYSTEM_PROMPT,
                      max_context=8192,
                      max_predict=512)
        tg_context.user_data[chat_id] = aisha
    aisha = tg_context.user_data[chat_id]
    aisha.load_context(chat_id)
    return aisha

async def process_message(update: Update, context: ContextTypes.DEFAULT_TYPE):
    chat_id = update.effective_chat.id
    user_id = str(update.message.from_user.id)
    aisha = get_aisha(str(chat_id), context)
    aisha.add_user_request(update.message.text,
                           system_injection=SYSTEM_INJECTION.replace("{user_id}", user_id))
    text_response = aisha.completion(temp=0.0, top_p=0.5)
    aisha.save_context(chat_id)
    if "ignoring_message" not in text_response:
        await context.bot.send_message(chat_id=chat_id,
                                       text=text_response,
                                       reply_to_message_id=update.message.message_id)

application = Application.builder().token(TG_TOKEN).build()
application.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, process_message))
application.run_polling()

Chainlit example

import chainlit as cl
from aishalib.aishalib import Aisha

@cl.on_chat_start
async def on_chat_start():
    aisha = Aisha("http://127.0.0.1:8000/completion",
                  "CohereForAI/c4ai-command-r-v01",
                  prompt="Ты отвечаешь на русском языке.",
                  max_context=4096,
                  max_predict=512)
    cl.user_session.set("aisha", aisha)

@cl.on_message
async def on_message(input_msg: cl.Message):
    output_msg = cl.Message(content="")
    await output_msg.send()
    aisha = cl.user_session.get("aisha")
    aisha.add_user_request(input_msg.content)
    response = await cl.make_async(aisha.completion)(temp=0.5, top_p=0.5)
    output_msg.content = response
    await output_msg.update()

Document search example

from aisha import Aisha

MODEL_ID = "microsoft/Phi-3-medium-128k-instruct"
COMPLETION_URL = "http://127.0.0.1:8000/completion"

with open("document.txt") as f:
    text = f.read()

prompt = f"""Ты - поисковая система.
Ниже представлен следующий текст по которому необходимо выполнять поиск.
## Текст:
{text}
"""

aisha = Aisha(MODEL_ID, COMPLETION_URL, prompt=prompt, max_context=40000, max_predict=8192)
user_request = "Вопрос по документу"
prompt = f"Верни релевантные фрагменты текста, которые наиболее точно соответствуют запросу пользователя: {user_request}."
aisha.add_user_request(prompt)
print(aisha.completion(temp=0.0, top_p=0.0))

Run Llama.CPP Server backend

llama.cpp/build/bin/server -m model_q5_k_m.gguf -ngl 99 -fa -c 4096 --host 0.0.0.0 --port 8000

Install CUDA toolkit for Llama.cpp compilation

Please note that the toolkit version must match the driver version. The driver version can be found using the nvidia-smi command. Аor example, to install toolkit for CUDA 12.2 you need to run the following commands:

CUDA_TOOLKIT_VERSION=12-2
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update
sudo apt -y install cuda-toolkit-${CUDA_TOOLKIT_VERSION}
echo -e '
export CUDA_HOME=/usr/local/cuda
export PATH=${CUDA_HOME}/bin:${PATH}
export LD_LIBRARY_PATH=${CUDA_HOME}/lib64:$LD_LIBRARY_PATH
' >> ~/.bashrc

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

aishalib-0.0.5.tar.gz (6.2 kB view details)

Uploaded Source

Built Distribution

aishalib-0.0.5-py3-none-any.whl (6.3 kB view details)

Uploaded Python 3

File details

Details for the file aishalib-0.0.5.tar.gz.

File metadata

  • Download URL: aishalib-0.0.5.tar.gz
  • Upload date:
  • Size: 6.2 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.0 CPython/3.10.12

File hashes

Hashes for aishalib-0.0.5.tar.gz
Algorithm Hash digest
SHA256 129e91d5aab20fb983dc571e672ce9e3be2d4dc28eac74e74885018d51cb68f0
MD5 21198af12e9b11105a78a4e547ffc30e
BLAKE2b-256 432a2eda4770779cd18f7448d06dd4edb8d916bfe7371d450b5224c5fc14b5b3

See more details on using hashes here.

File details

Details for the file aishalib-0.0.5-py3-none-any.whl.

File metadata

  • Download URL: aishalib-0.0.5-py3-none-any.whl
  • Upload date:
  • Size: 6.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.0 CPython/3.10.12

File hashes

Hashes for aishalib-0.0.5-py3-none-any.whl
Algorithm Hash digest
SHA256 f9ff40d5580cf4015acac2c3fca6636711ec118da40abd8bf0871fb7aa5a0982
MD5 93ec5c3fb4387502d39fd212058a0ce9
BLAKE2b-256 839d095f78167c36d30d7906e44b7b04f7b8591f5c3aabff691c0b0acd9b087b

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page