[dchest]

You can use OpenLLaMA models (Apache 2.0 license, unrelated to LLaMA apart from their architecture and general approach to training):

https://huggingface.co/TheBloke/open-llama-7b-open-instruct-...

https://huggingface.co/SlyEcho/open_llama_7b_ggml

https://huggingface.co/SlyEcho/open_llama_3b_ggml