documentation/docs/tooltips/experiments/_evaluate-before-training.mdx

This option lets you evaluate the model before training, which can help you judge the quality of the LLM backbone before fine-tuning.