AddOns
Collection
28 items
•
Updated
interchangable !!!
Actually my Got to Lora Quite large training cycle so when applying to merged misrals it does realign them . i usually apply this lora to each model premerge , so the outcome will also be merged with this lora , and recieve some traing with dolphin databases ; hopefully still alligned to the dataset... so multi layered dolphin data is injected in to the hosts . to give a woarm start : This is applied to smaller layered models also ! (as it generally has some effect especially on the raw base models .... something is in there working!)
This mistral model was trained 2x faster with Unsloth and Huggingface's TRL library.