Re:Gptee: LLMs done the Unix-y way
by zurawiki on Thursday March 02, 2023 @11:01PM
(#35005648)
The idea behind gptee is to bring composability to language models at the shell script level. In line with the unix philosophy, we simplify the LLM interface to just stdin and stdout
Re:Gptee: LLMs done the Unix-y way
by compressedgas on Thursday March 02, 2023 @11:09PM
(#35005708)
Please specify in the README near the top that this uses OpenAI's API and does not run the model locally. I thought this was neat until I looked to see if it used the API.
Re:Gptee: LLMs done the Unix-y way
by smoldesu on Thursday March 02, 2023 @11:44PM
(#35005918)
I threw something like this together, but it probably won't win any awards. Give it a whirl if you have cargo/libtorch installed: https://github.com/toasterrepairman/qubert
It's not even close to ChatGPT intelligence (HN was sure to remind me of that) but I think the results are pretty good for minimal-applied-effort and minimal resource usage. It's been my starting place for AI POCs recently (some of which I switch to OpenAI for). The current pruned models are extremely dumb, but I have faith they will improve.
https://en.wikipedia.org/wiki/Unix_philosophy
It's not even close to ChatGPT intelligence (HN was sure to remind me of that) but I think the results are pretty good for minimal-applied-effort and minimal resource usage. It's been my starting place for AI POCs recently (some of which I switch to OpenAI for). The current pruned models are extremely dumb, but I have faith they will improve.