Add bleu, rouge evaluation indicators.

[LianxinGao]

we need model evaluation method.

[mikecovlee]

I have wrote a BERT based auto evaluator mainly focus on fast evaluation to checkpoints of LLM for early stop when fine-tunning.

[LianxinGao]

I have wrote a BERT based auto evaluator mainly focus on fast evaluation to checkpoints of LLM for early stop when fine-tunning.

OK，maybe the way below is a necessary method to offer

python mlora.py \
  --eval \
  --base_model {model_path}  \ 
  --checkpoint {fine-tuned checkpoint path} \
  --dataset_path {dataset to eval}