[Great Model]
First of all, i liked your method of model finetuning. this thing came to my mind too upscaling a model then finetuning on the duplicate layer to prevent any data loss and make the model better than the base one. I used this technique for my model [OpenCrystal-L3-12B] using llama3 12b as base then finetuning it.
Good things :
- it's good at roleplaying
- remembers small details in my expirence
- better at nsfw
- good in instruct following
Templates :
(i used alpaca fromat) it feels alive, and for normal tasks i use the mistral format (mistral format seems bland and dry) for chatml (it's not as good as alpaca)
EDIT : Alpaca preset is the best one so far.
@Darkknight535 I'm actually surprised Alpaca worked for you in RP. Does the AI tend to speak for you?
@Darkknight535 I'm actually surprised Alpaca worked for you in RP. Does the AI tend to speak for you?
nope.