Title. Really cool work but I am kinda curious: how did you guys convert phi-3 mini into the mistral architecture? Do you mind writing a quick blog post or share a script or something similar.
I am also hesitant because idk if the lora adpters produced by unsloth trained on phi-3 mini can be used on the original phi-3 mini... or if im forced to always use unsloth's version during inference
