Current Status on SnooSpoof (TODO)

[cnshing]

Implemented

• Basic parsing for generated text
• Basic scrapper
• Finish Unit Testing on Parsing

Todo

• Add Figma, Illustrator design files to repository
• Add check for parsing invalid texts
• Finish Unit Testing on Scrapper
• Implement Flask Middleman API to use every component
• Implement Trainer Component
• Implement Encoder Component
• Create adapter interface between front-end and backend
• Create React components from Figma design
• Abstract the Scrapper component to be injectable with any PRAW Reddit Instance
• Optimize model/text pipeline mechanisms for repeat usage
• Basic data validation for encoder and dataset
• Remove gentext2json as it will most likely not be used in the future
• Remove trainer_unittest as it is unneccesary to test a high level abstracted component of a heavily tested library
• Compress the middleman generate API query parameter url to prevent it from exceeding 2048 characters

Nice to Have

• Allow any component to be modularly run on seperate instances and configurations(Google Colab?)
• Figure out some way for SnooSpoof to be hosted indepedently/freely
• Add a caching system to optimize scrapper/dataset
• More robust validiation for datasets (pydantic?)
• Refactor requires function to not have to call itself to verify, rather let it automatically verify the features automatically whenever the requires decorator was applied
• Refactor tag format to make it customizeable [prompt] or "prompt: " or (prompt), etc
• Make each tag format more disconnected from the underlying text samples as to not pollute the data, by marking each tag format as a special token, making each token less distinguishable from regular text, etc
• Allow asynchronous processing of text generation
• Make Google Colab FastAPI notebook public
• Containerize application as docker container for easy deployment for GPU renting servers