Yeah, I mention this in the post but this variant of LLaMA isn't storing any of the conversation in memory so it doesn't have context on the prior questions. You're starting fresh with each prompt. We have some ideas for how to improve this though... more soon :)
