# How to evaluate bert model from MLM task result?

**URL:** <https://discuss.huggingface.co/t/how-to-evaluate-bert-model-from-mlm-task-result/57375>\
**Category:** Beginners\
**Created:** [October 4, 2023, 1:16pm UTC](https://discuss.huggingface.co/t/how-to-evaluate-bert-model-from-mlm-task-result/57375 "2023-10-04T13:16:36Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![JackieShi](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/jackieshi/32/19863_2.png) [@JackieShi](https://discuss.huggingface.co/u/JackieShi)\
**Post date:** [October 4, 2023, 1:16pm UTC](https://discuss.huggingface.co/t/how-to-evaluate-bert-model-from-mlm-task-result/57375/1 "2023-10-04T13:16:36Z")

</div>

Now, I have to train BERT from scratch using my own domain data. And in my situation, I don’t have to do NSP, instead only MLM task is enough.  
I’ve trained many times and the loss is always around 2 ~ 4. But I don’t know the “good” value for this task. So I wonder that is there any way to evaluate or is there any standard value which can be a reference?
