???????????? ?????????????????? ???????????????????? ?????????? ?????????????????????????????? ??????????????????????? ????????????????????????? ???????????????????????????? ?????? ??????????????????????? ?????????????????????????? ??????????????????????????? ?????????????????????????????? ????????

You may have to utilize the gpu_memory_limit and/or lora_on_cpu config possibilities to stop running away from memory. If you still run from CUDA memory, you'll be able to attempt to merge in procedure RAM with Posted in?? ???imToken??? ???imToken???imToken?????imToken????????????????imToken???????