Llama 3 is out

[Galunid]

https://ai.meta.com/blog/meta-llama-3/
https://llama.meta.com/llama-downloads

[mirek190]

..and is insane good ...

[stewartoallen]

Even building llama.cpp from today's source with the Llama 3 patches, when using most Llama 3 models (with main), it's still wandering off into gibberish land after generating a few hundred tokens. Using the same prompt and same Llama 3 models in something like LM Studio (which uses llama.cpp), I get coherent results. What am I missing?