2 ms·
It seems to perform best with 16K context, but works well with up to 32K context. This is interesting, especially coming from Amazon, as vanilla Mistral-Instru
by brucethemoose2 3y ago
It seems to perform best with 16K context, but works well with up to 32K context.
This is interesting, especially coming from Amazon, as vanilla Mistral-Instruct is much better than llamav2 7B, and extremely fast on modest hardware.