Send nice texts to your friends using LLMs
Project description
compLLMents
Description
This package enables you to send scheduled, uplifting, multi-modal AI-generated text messages to your friends. (Though they won't remain friends long if your only communicate is automated 😉)
It works by first using an LLM to generate a batch of positive and complimentary messages in the language of your choice. Then, a multilingual sentiment classifier scores all the generated posts and selects the most positive to send either as an SMS or over WhatsApp. Further, after recording a few minutes of audio, a custom text-to-speech model will record the message in your voice. Here is the accompanying Colab notebook.
Table of Contents
Installation
First, ensure that poetry
is installed.
poetry install
poe install-pytorch
To download files to store locally and save time of future downloads, run:
download -m path/on/huggingface
SMS
First create a free Twilio account and create a phone number (note: Twilio automatically prepends the message Sent from your Twilio trial account
to free-tier accounts). Copy your credentials from the dashboard into the TWILIO_CONFIG
dictionary in config.py
. An example config will look like:
{
"account_sid": "a_string",
"auth_token": "a_token",
"from_": "+11234567890",
}
You must log in from your computer for the messages to send.
Audio
Training a custom test-to-speech (TTS) model requires a corpus of recordings to fine tine on. The Mozilla Common Voice initiative is a crowdsourced voice dataset. After creating an account, you can record yourself speaking sentences in the language of your choosing. Once finished recording, go to Profile
>> Download My Data
and copy the URLs you see into the MOZILLA_CONFIG
dictionary in config.py
like below:
MOZILLA_CONFIG = {
"first_url": "",
"second_url": "",
}
Usage
Texts are sent by running:
send -r recipient-name -s sender-name -n +11234567890 -l language -t type -b -sa
send --help
explains the parameter options. Pass your OpenAI API key using -o
to use their models.
You can send custom messages by chaning the text in the TEMPLATE
object in main.py
You can set custom model configuration in the INFERENCE_CONFIG
object in conifg.py
including swapping out models, increasing the output length by chaning max_new_tokens
or increasing the randomness in reponses by raising temperature
or top_p
. The default language generation model is NousResearch/Nous-Hermes-13b
which is the best performing open-sourced LLM at the time of creation. The default sentiment analysis model is cardiffnlp/xlm-roberta-base-sentiment-multilingual
which supports 8 languagees: arabic
, english
, french
, german
, hindi
, italian
, portuguese
, and, spanish
.
To schedule texts to be sent at regular intervals, create a crontab similar to the example in cron
.
Examples
Here are some examples messages and their sentiment score.
Message | Sentiment Score |
---|---|
Hey Austin! Just wanted to remind you that you are an amazing friend and such a positive force in my life. Keep being you, because you're pretty darn great. | 0.939329206943512 |
Hey Austin! Just wanted to let you know that you're an amazing friend and I'm lucky to have you in my life. Keep being your awesome self and never forget how much you're loved and appreciated! 😊 | 0.9417279958724976 |
Hey Austin! Just wanted to remind you that you are an amazing friend and person. Your kindness and positivity always brings a smile to my face. Keep being you, because you're awesome! :) | 0.946333646774292 |
The final recording is below. This was generated by fine-tuning the text-to-speech model on 200 sentences. The more data it is given, the more life-like it will sound.
Tests
Forthcoming...
Releases
0.1.0
- Self-hosted or OpenAI generated compliments
- SMS and WhatsApp supports
- Sentiment-based message selection
- Message scheduling
- User-friendly Colab notebook
0.1.1
- Fixed
ReadME.md
0.2.0
- Custom voice messages
- Birthday and custom message templates
License
MIT License
Copyright (c) [2023] [Austin Botelho]
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for compllments-0.2.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | a5df2f14eaca2dada09aba6eb4544ad6c513fbd31db42494562defa1510ce361 |
|
MD5 | 7c333c7bba3c046b1e3ea92e21a7bee2 |
|
BLAKE2b-256 | d737ea2f238830b95b86e38624a507c09d2532f51901371d7a85d0e6ca2c56cd |