Load and run the model:

vllm serve "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit"
but:
KeyError: 'layers.0.mlp.down_proj.weight'
Loading safetensors checkpoint shards: 0% Completed | 0/1 [00:10<?, ?it/s]