this post was submitted on 19 Apr 2022
3 points (100.0% liked)

Telegram

1076 readers
4 users here now

A community about Telegram, the popular messaging app.

Rules:

  1. be kind and avoid aggressive or offensive language toward other people
  2. do not submit links that are aimed at advertising your channels, groups or bots
  3. English only

founded 5 years ago
MODERATORS
 
you are viewing a single comment's thread
view the rest of the comments
[โ€“] [email protected] 0 points 2 years ago (1 children)

Fun fun fun. Where's the source code comrade? :)

Also, can you maybe tell us more about what data set you trained your model with?

[โ€“] [email protected] 1 points 2 years ago

ty ๐Ÿค—

the code for the bot isn't yet ready, because i just started the project, and i'd be comfortable releasing it once i added some more functionality and customisability, around the moment i'd be making the bot public

and there's one issue i have to overcome before making the bot itself public: the computer i'm running this bot on is kinda old (like 2010), and it has a gpu that doesn't support recent versions of cuda, which is the gpu compute library used by one of the libraries in this project, therefore all the inference is run on the CPU, which is slow and very computationally intensive

i'm not sure what to host this on: i'll probably wait until the GPU prices have come down, and see if i can purchase some modern-ish GPU in order to run this on, bc renting so much GPU compute can cost a looot

i'm using a pre-trained model for this called BART, because training a good NLP transformer model of such magnitude (0.5 billion parameters) is an astronomical task not only in terms of computation, but also in terms of the required expertise, which i certainly do not possess at the moment (i just graduated high school lol)

let me know if you'd like to use something like this bot :)