unsloth

/

llama-3-8b-bnb-4bit

like 176

I need to train the model, but on multiple GPUs. I copied the code from github (I hoped it was automatically configured), it doesn't work. I asked people, they told me to use the ProcessGroup from accelerate, it gave me 0 in device_map, inserted it into get_peft_model and nothing - only one video card was loaded. Who knows what to do? Help!