Isn’t llama.cpp low level and highly optimized? There may not be that much overlap in the required skill sets.

This doesn’t seem correct to me. I saw in an Ollama issue mentioned in another comment, an Ollama contributor said:

> As you pointed out, we carry patches, although in general we try to upstream those.

— https://github.com/ollama/ollama/issues/2534#issuecomment-19...

So I followed the link to his profile and saw that he has opened some non-documentation pull requests for llama.cpp:

https://github.com/ggerganov/llama.cpp/pull/5244

https://github.com/ggerganov/llama.cpp/pull/5576

I didn’t dig any deeper, but it took me less than thirty seconds to find those so I expect there are more.

There’s no legal obligation.

…but I think it’s fair to say there’s a social obligation tip your hat to the shoulders you stand on.

This has come up before, and they still do exactly the same thing, and there absolutely zero chance they haven’t heard the critique about it.

So you must presume it’s a deliberate choice, rather than “oops we didn’t think of that”… /shrug

If you don’t want to be called out for it, don’t do it. I’m not particularly sympathetic to them in this case.

There is no license that obligates contributing upstream.