LMFlow icon indicating copy to clipboard operation
LMFlow copied to clipboard

Is it possible to evaluate while training

Open abc20220327 opened this issue 2 years ago • 2 comments

我看到 run_finetune_with_lora_save_aggregated_weights.sh中 有以下参数

--do_train
--do_eval
--evaluation_strategy "steps"
--eval_steps 1000
--eval_dataset_path ${eval_dataset_path} \

是否可以 修改 run_finetune_with_lora.sh 加入上述参数,从而实现一边训练一边评估

我尝试了下,抛以下错误:

image

能否解惑,谢谢!

abc20220327 avatar Jun 12 '23 07:06 abc20220327

需要你安装sklearn包。可以尝试安装一下 pip install scikit-learn==1.2.2

Dominic789654 avatar Jun 13 '23 11:06 Dominic789654

非常感谢,可以用了 @Dominic789654 ,不过我看到 之前有个 帖子 https://github.com/OptimalScale/LMFlow/issues/218 ,提到他们目前放弃了 这种 eval 的方案,可是我试用了下还是可以输出 验证集的Loss的,你知道为什么吗 @shizhediao

abc20220327 avatar Jun 13 '23 16:06 abc20220327

我们当前的代码支持了输出验证集loss的方案。之前提到的是旧的方案,可以忽略

shizhediao avatar Jul 10 '23 02:07 shizhediao

This issue has been marked as stale because it has not had recent activity. If you think this still needs to be addressed please feel free to reopen this issue. Thanks

shizhediao avatar Sep 30 '23 19:09 shizhediao