/aiyu

core shell functions building blocks for advanced AI pipelines

Core shell functions for AI

What is AIyu:

In essence, the Aiyu Shell pipelines serve as the interweaving adhesive that binds the various AI components together - a technological superglue of sorts!

Aiyu allows to build pipelines that easlily leverage the most exciting ai technologies using any laptop.

example

input: audio in some language; output: audio in French
m2a; a2a fr; a2sk

Listen to your microphone and outputs an audio file (m2a), then use the previously created audio and translate it to French (a2a fr) and lastly plays the audio to your speaker (a2sk).

input: audio question in some language; output: audio GPT response in Italian
m2a; a2p; p2a; a2a it

Listen to your microphone and outputs an audio file (m2a), then use the created audio and transcribes it to textual prompt (a2p), use the prompt to query GPT3.5-turbo and produce an audio of the answare (p2a) and lastly, use the created audio and translate it to Italian (a2a it).

Glossary of abbreviations

Inputs and Outputs

c  -> code     ex: sourcecode of a python program
p  -> prompt     ex: "how can I escape the matrix?" 
t  -> text       ex: .txt file of a motivation letter
s  -> subtitle   ex: .srt file of a movie subtitles
a  -> audio      ex: .mp3 file of a recorded conference 
                 I/O: {sk -> speaker, m <- microphone}

Current state of implemented functions

prompt text subtitle audio code
prompt p2p p2t ap2s * p2a p2c
text t2p t2t at2s * t2a p2c
subtitle - s2t s2s s2a -
audio a2p a2t a2s a2a a2c
code - - - - c2c

Features:

  • Hackable
  • It support most languages
  • Linux and Mac compatible. Windows is coming soon.
  • Beautiful Charm TUI interface written in Go.
  • Bash function editor and experimental Markdown reader.
  • focused on CPU performance
  • Documented (very soon)
Concise pipelines (inputs2output)

🔹 Ask gpt and gtts answer to speaker (text2speaker) -> t2sk
🔹 Ask gpt and gtts answer to audio (text2audio) -> t2a
🔹 Ask gpt and produce specifically code (code2text) -> c2t
🔹 Take audio and produce text transcript (audio2text) -> a2t
🔹 Take audio and produce subtitles transcript (audio2subtitles) -> a2s
🔹 Prompt gpt and produce text (prompt2text) -> p2t
🔹 Take text and translate into text (text2text) -> t2tr
🔹 Take audio and enhance quality into audio (audio2audio) -> a2a
🔹 Take audio & text transcript and produce subtitles (audio+text2subtitles) -> at2s
🔹 Take text and stable diffusion produce image (text2image) -> t2i

Screenshots & Tutorial: Wiki

In the terminal:

download directory and enter it:
git clone https://github.com/GabrieleRisso/aiyu.git 
cd aiyu/
set the key to global scope:
. env.list
load a function:
. t2s
use it like this:
t2a "How are you ?"
Dependencies

They are prompted if missing and a link to install them is provided. Functions mostly leverages Python libs installed via pip and pre-build Docker images. Memory usage statistics will be out soon.

Who's project is doing the AI heavy lifting?

Open for collaboartions; let's make aiyu awesome toghter

  • per-command tuning
  • wiki documentation
  • suggestions are desired!

Citation

If you utilize this reposistory please consider citing it with:

@misc{aiyu,
  author = {Gabriele Risso},
  title = {aiyu: core shell functions for advanced ai},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/gabrielerisso/aiyu}},
}

Copyright

Copyright © 2023 Gabriele Risso.

Contact for custom implementation:

gabriele.risso@protonmail.ch