Llama 3 models are pretty spectacular

[dmm]

Both 70B and 8B versions are very impressive on initial blind comparisons: chat.lmsys.org/?leaderboard

If Llama 3-70B-Instruct turns out to be indeed more or less equivalent to early GPT-4, this would have a lot of wide-ranging implications.

One can use Llama 3-70B-Instruct at www.meta.ai/ for free.

Comments

[dmm]

There is a lot of things which are interesting about this model.

For example, it is a usual old-fashioned decoder-only Transformer, not a GPT-4-like "Mixture-of-Experts", which means that it should be much easier to fine-tune.

[dmm]

https://news.ycombinator.com/item?id=40077533

[dmm]

It allowed me to define and use a simple custom syntax, just like GPT-4:

https://github.com/anhinga/with-GPT-4/blob/main/Llama-3-comparison/custom-syntax.md

So, it passed my first software engineering test just fine.