在使用examples/inference/evaluate.sh脚本评测lora模型时，评测结果都是0

[lwj2001]

Reminder

• [X] I have read the README and searched the existing issues.

Reproduction

我的evaluate.sh脚本如下：

CUDA_VISIBLE_DEVICES=0,1 llamafactory-cli eval \
    --model_name_or_path /Data3/liwenjie/model/HIT-SCIR/Chinese-Mixtral-8x7B \
    --adapter_name_or_path /Data3/liwenjie/ft_model/Chinese_mixtral/lora/sft_3epoch \
    --template fewshot \
    --finetuning_type lora \
    --task_dir /Data1/home/fanziqi/LLaMA-Factory/evaluation \
    --task ceval \
    --split test \
    --save_dir /Data3/liwenjie/eval/Chinese-Mixtral-sft_3epoch/lora/predict \
    --lang zh \
    --n_shot 5 \
    --batch_size 4

运行的结果是：

        Average: 0.00                                                                                                                                                                                                          
           STEM: 0.00
Social Sciences: 0.00
     Humanities: 0.00
          Other: 0.00

Expected behavior

No response

System Info

No response

Others

No response

[codemayq]

删除一下 lora相关参数，先做一下原模型的评测试试

[codemayq]

CUDA_VISIBLE_DEVICES=0 llamafactory-cli eval --model_name_or_path /media/codingma/LLM/llama3/Meta-Llama-3-8B-Instruct --template llama3 --task ceval --split validation --lang zh --n_shot 5 --batch_size 1

结果如下

        Average: 53.86                                                                                                                                                                              
           STEM: 52.79
Social Sciences: 60.73
     Humanities: 52.53
          Other: 51.04

[hiyouga]

ceval 没有 test 标签，只能用 --split validation