Gonna make an NLP model that'll respond to API calls. Maybe make it into a Twitter bot too or something.. idk
- get scripts
- get better formatted scripts for NLP
Show: https://transcripts.foreverdreaming.org/viewforum.phpf=1721
Important quotes: https://en.wikiquote.org/wiki/JoJo%27s_Bizarre_Adventure
Manga: picture-to-text scanner maybe?
Manga: https://mangadex.org/title/136/jojo-s-bizarre-adventure-part-1-phantom-blood
- Make REST API that will call model when prompted and return quote when unprompted
- Flask is lame, learn Rust and use Axum?
- Fix output not rendering newlines
- Figure out how to use Jojo-GPT2 in Rust
-
Understand NLP basics
-
Train jojo-gpt2
-
Train a better one
-
Test different models Llama 2, GPT-NeoX, GPT-J, Vicuna-13B, XGen-7B
-
Understand tokenization, stemming, and lemmatization
-
Apply/practice concepts
Domain: https://instantdomains.com/search?q=jojosbizarreapi ITS MINE
https://www.heroku.com/
https://www.pythonanywhere.com/
https://www.digitalocean.com/
https://www.shuttle.rs/
AWS Lambda?
Digital Ocean Web App 👎BOO tomato tomato tomato
Digital Ocean Droplet
- Build and containerize the API
- Create CI/CD pipeline with GitHub Actions
- Deploy API and use domain
- HTTPS
- Manually upload model through SSH or store on Google Drive lol
Flask:
python flask_api.py
Axum:
If you want to use pytorch instead of libtorch, you may need to follow these instructions:
pip install torch torchvision torchaudio
python -m venv C:\Users\Rybeardawg1\AppData\Local\Programs\Python\Python39\Lib\venv
set LIBTORCH_USE_PYTORCH=1
As of now, tch version 0.14.0 crate supports up to PyTorch 2.1.0. If you need to downgrade, use pip install torch==2.1.0
If you want to use libtorch, download here:https://pytorch.org/ and you may need to set these env variables:
set LIBTORCH_USE_PYTORCH=0
set LIBTORCH_BYPASS_VERSION_CHECK=0
cargo build
cargo run