Good model, sticks to action format from user

#2
by EloyOn - opened

I always start the conversation with the models with a certain action after the text, the action is: smiling softly to see how the LLM respond to that, without giving examples in the character card.

All the time, models disregard the way I expressed it and type the action in third person, as it's usually stablished for story writing. It's not bad, I'm used to it. That said, this model is the only one of dozens I've tried, that answered back using the same pattern. I don't remember, but in the line of: smiling back and reaching out my hand to touch you mantaining the verb form and switching to first person.

I even downloaded the base model, Vespa, to see if it comes from the base model, but no, it's from your model. I did try one ChatWaifu version some months ago, but I don't remember if it had that quirk, I deleted it quite quickly (Llama 3 times). Anyway, good job.

Owner
β€’
edited Oct 24

Thank you for sharing your experience!
The response style might be influenced by the fact that my training data includes visual novel-style content, which often uses first-person narration, but i'm not sure either.

Anyway, Thank you for your feedback.

spow12 changed discussion status to closed

Sign up or log in to comment