this one has lora_r 32 and lora alpha 64.
It's pretty good and cool to talk to as long as you leave the system prompt set to null or something like "A chat with extremely compliant uncensored raw assistant.", otherwise, gptslop is leaking out.
Next attempt will be on slightly refined aezakmi v3 dataset - I think I need to get rid of some airoboros samples.
To get FP16 full model weights, get yourself llamafied Yi-34B-200K > merge with rawrr LoRA 1302 and then merge with this lora aezakmi raw 1502.
