Impressive
Hi pot99rta,
just wanted to say thank you.
The q6 is a very impressive model, few logical errors in conversations (significantly less than common with other current Models), feels very stable overall, has barely any flaws, (q5 has some more, like breaking out of a story or chat with system messages, at least in an i1 gguf I tested), good in understanding circumstances and adapting to them. Wenn you reach the 8196 Mark, it tends to forget some details or Instructions from the very early part of the conversation, including the system prompt. But even this feels significantly less inconvenient, than the usual forgetfulness of most other models in the range between 7B and 24B, so it's still excellent and the best I have tried out so far.
Greetings,
Huegli