some people at stanford finetuned the smallest of these models to
follow instructions like openai’s models
their code is at https://github.com/tatsu-lab/stanford_alpaca . the
finetuned model itself is not shared but people are replicating it
using the code. a web demo is at
https://www.reddit.com/r/MachineLearning/comments/11h3p2x/d_facebooks_llama_leaks_via_torrent_file_in_pr/
Just FYI, it’s really easy to get legitimate access. All I did was put
down that I’m a student studying machine learning and wanted to test
the model, no proof required. Got access in a few
https://twitter.com/Teknium1/status/1631322496388722689
https://t.co/QwqDmpdftO
magnet:?xt=urn:btih:b8287ebfa04f879b048d4d4404108cf3e8014352=LLaMA
(also untested)
online interface? https://huggingface.co/spaces/chansung/LLaMA-7B
https://news.ycombinator.com/item?id=35007978 [says it links to
this is a cutting edge public language model that was kept private to
the academic community. if i recall correctly its major feature is
significant downsizing via use of much more training data. it is
likely more powerful than common for-pay models such as those sold by
openai.
i have not
