On a A100 40GB version, you could train Facebook's 1B model over a day or few. Exploit-LLM1 only has 46M and Exploit-LLM2 will have around 200M, and only Exploit-LLM3 or Exploit-LLM4 will take as long as Facebook's smallest model. Yet Exploit-LLM1 already provides usable insights if you were seriously that desperate; please wait until 2, 3, 4, etc.

Exploit-LLM 1, trained for $3 on Colab

If you're looking at development concept test Terminal, it doesn't let you relabel if you misclick easily; however the deployment version will have web UI. Mocking Google Marketing model; I use the full html and a backend with frontend persistence support, on the deployable version.

Gmail API PersonalModel

PDFSage completed version of StupidityModel - compare this performance with yours

StupidityModel - Completed

Incomplete StupidityModel, for you! used for corporate training, etc. Run inference on CPU for free, sign up $10/month to use NVIDIA Tesla T4 GPUs

HateCrimePreventionModel

Put some flowers in your Google Drive - you need a Kaggle account for this version, and their API json in your Google Drive first

PDFSage News and Updates

Exploit-LLM 1, trained for $3 on Colab

Gmail API PersonalModel

StupidityModel - Completed