I noticed there's a couple of open issues on llama.cpp investigating quality issues. It's interesting if a wrong implementation still generates plausible output

I noticed there's a couple of open issues on llama.cpp investigating quality issues. It's interesting if a wrong implementation still generates plausible output. It sounds like an objective quality metric would help track down issues. https://github.com/ggerganov/llama.cpp/issues/129 https://github.com/ggerganov/llama.cpp/issues/129 https://github.com/ggerganov/llama.cpp/issues/173 https://github.com/ggerganov/llama.cpp/issues/173