nice one.

i would add the natively trained alpaca models, which exist besides alpaca-lora. see my model card for this:

https://github.com/underlines/awesome-marketing-datascience/blob/master/llama.md#3rd-party-llama-and-alpaca-models

and here's an overview of almost every LLM under the sun:

https://docs.google.com/spreadsheets/d/1O5KVQW1Hx5ZAkcg8AIRjbQLQzx2wVaLl0SqUu-ir9Fs/edit#gid=1158069878

Update: Found LAION-AI/OPEN-ASSISTANT a very promising project opensourcing the idea of chatGPT. video here

They have two versions: public and azure.

no need to integrate the ads into the LLM. Just integrate it into the UI that users use to converse with the AI. Between Answers you can either inject ads, or you can alter answers to contain certain brands.

Very unethical, and that's why I hope this becomes detached from big corps like OpenAI that do this behind a locked down API...

DeepSpeed-MII supports Bloom int8 and fp16.

brands / advertisers pay money to the LLM platform to run highly targeted ads in the LLM interface (for example chatGPT, lawGPT, medGPT etc.) the LLM platform pays a share of that adrevenue to content creators, that it uses for training and finetuning.