lm-eval run --model hf --model_args pretrained=meta-llama/Llama-3.1-8B-Instruct --tasks hellaswag --limit 100

Evaluates only the first 100 hellaswag examples, cutting a long run down to minutes so you can verify the setup before committing to the full benchmark. The limit accepts a fraction too, like 0.1. Remove --limit for the real, complete evaluation.