OpenAssistant

/

oasst-sft-6-llama-30b-xor

like 946

Pythia can be used in commercial. But not for Lama, there is nothing we can do, because it is a product of Meta, and they restrict such cases.

But Pythia ones are also not commercially viable in my view, because they still use alpaca data or vicuna data, which includes a bunch of chatGPT output, and their ToS are clear:

https://openai.com/policies/terms-of-use

Restrictions. You may not ... use output from the Services to develop models that compete with OpenAI;

At least at H2O.ai we have stayed away from any ShareGPT, alpaca data, explicitly to avoid such issues with our 7, 12, and 20B neox models:

https://github.com/h2oai/h2ogpt