Sorry to ping but I'm still having no luck with this and can't find any info anywhere. I seem to be alone in this and I can't figure out why.
#Issues with Wizard mega 13B
echo hound
echo hound
shadow flame
That looks like pretty reasonable output to me for a 13B model...
echo hound
Using Snoozy and Vicuna 13b they can both modify previous responses.
This one seems to have no context of what's been said, as you can see in that last screenshot. It's completely ignoring the prior conversation and writing code completely unrelated to anything that's been said.
The other models are usually "professional", understanding that they're AI assistants and being eager to help, but Wizard Mega seems to ignore it's character and make up completely random roles for itself.
shadow flame
If these are your complaints, you need to take them to the authors of the model. This is working as intended.